Re: snmp process monitoring....
Can someone tell me how to exit the list while I am on vacation. I want to set an away message and don't want to flood the list with junk. Joe Temple IBM Distinguished Engineer Sr. Certified IT Specialist [EMAIL PROTECTED] 845-435-6301 295/6301 cell 914-706-5211 Home office 845-338-1448 Home 845-338-8794 -- For LINUX-390 subscribe / signoff / archive access instructions, send email to [EMAIL PROTECTED] with the message: INFO LINUX-390 or visit http://www.marist.edu/htbin/wlvindex?LINUX-390
Re: Who's been reading our list...
The shared L2 reduces the penalty for those situations when you can't avoid dispatching on a new engine. That is when the system is very busy. This is one of the reasons for the difference in utilization. As the machine gets busier other machines are forced into L2-L2 or remote L3-localL1 (victim cache) transfers which have a high penalty. In z the migration is from shared L2 to L1. The less affinity scheduling delays dispatching, the more the system behaves like a multiple server single queue system, which is the optimum case. The more scheduling delays dispatching the more the system behaves like multiple single server single queue systems, which will not perform well if the load has skew or high variability. Thus if the affinities are hardened (often done in skewless benchmark runs) skew will cause some cpus to overload while others are idle. If there is no affinity then there are more cache migrations. In between the there is a combination of the first and second cases and it is a matter of what the migration penalty is v the queueing penalty for affinity scheduling. Of course this is yet another reason that relative capacity is workload dependent. Another aspect of z's common L2 is that it always holds a copy of the data in the L1s attached to it and therefore snooping is avoided. High end System x systems (X460 class) do this by keeping a shadow directory that covers the on chip caches. Joe Temple Distinguished Engineer Sr. Certified IT Specialist [EMAIL PROTECTED] 845-435-6301 295/6301 cell 914-706-5211 Home office 845-338-1448 Home 845-338-8794 Alan Altmark/Endicott/ [EMAIL PROTECTED] To Sent by: Linux on LINUX-390@VM.MARIST.EDU 390 Port cc [EMAIL PROTECTED] IST.EDU Subject Re: Who's been reading our list... 05/18/2006 08:55 AM Please respond to Linux on 390 Port [EMAIL PROTECTED] IST.EDU On Thursday, 05/18/2006 at 10:03 ZE2, Martin Schwidefsky [EMAIL PROTECTED] wrote: The cache is a different story. Mainframes have the advantage of a shared level 2 cache compared to x86. If a process migrates from one processor to another, the cache lines of the process just have to be loaded from level 2 cache to level 1 cache again before they can be accessed. On x86 it goes over memory. The cache designs on the mainframe change from generation to generation to deal with more work, changes in the relationship of CPU speed to memory speed, and more CPUs. You want the benefits of cache, but you want to minimize the serialization/synchronization effects on the processors. This is why we do our best to dispatch a virtual machine on the same CPU as was used in the previous time slice. The relationship between the CPUs and a particular cache is not always equal, but is always the best if you use the same CPU again. Alan Altmark z/VM Development IBM Endicott -- For LINUX-390 subscribe / signoff / archive access instructions, send email to [EMAIL PROTECTED] with the message: INFO LINUX-390 or visit http://www.marist.edu/htbin/wlvindex?LINUX-390 -- For LINUX-390 subscribe / signoff / archive access instructions, send email to [EMAIL PROTECTED] with the message: INFO LINUX-390 or visit http://www.marist.edu/htbin/wlvindex?LINUX-390
Re: Fw: [LINUX-390] Who's been reading our list...
Yes tagging works, but you will find that the system z holds a lot more translations in a two tiered TLB and has tagging as well. Thus the System z does not have to retranslate as often. Joe Temple Distinguished Engineer Sr. Certified IT Specialist [EMAIL PROTECTED] 845-435-6301 295/6301 cell 914-706-5211 Home office 845-338-1448 Home 845-338-8794 Alan Cox [EMAIL PROTECTED] u.org.uk To Sent by: Linux on LINUX-390@VM.MARIST.EDU 390 Port cc [EMAIL PROTECTED] IST.EDU Subject Re: Fw: [LINUX-390] Who's been reading our list... 05/18/2006 07:23 AM Please respond to Linux on 390 Port [EMAIL PROTECTED] IST.EDU On Iau, 2006-05-18 at 10:03 +0200, Martin Schwidefsky wrote: On x86 it is the translation-lookaside-buffers (TLBs) which get flushed each time the control register 1 is loaded. Switching between threads is [%cr3 not 1 but thats by the way] fine because the use the same translation table. Switching between processes has a performance penalty. On mainframes the TLBs are not flushed for any context switch. Not on a current AMD x86 processor. The opteron and AMD64 processor line uses tags on the tlb entries so that it can avoid this without the underlying OS being changed. The cache is a different story. Mainframes have the advantage of a shared level 2 cache compared to x86. If a process migrates from one processor to another, the cache lines of the process just have to be loaded from level 2 cache to level 1 cache again before they can be accessed. On x86 it goes over memory. Long ago yes but even with private L2 caches (which have advantages too) you can send lines from cache to cache directly if your bus protocol is sane. Alan -- For LINUX-390 subscribe / signoff / archive access instructions, send email to [EMAIL PROTECTED] with the message: INFO LINUX-390 or visit http://www.marist.edu/htbin/wlvindex?LINUX-390 -- For LINUX-390 subscribe / signoff / archive access instructions, send email to [EMAIL PROTECTED] with the message: INFO LINUX-390 or visit http://www.marist.edu/htbin/wlvindex?LINUX-390
Re: Fw: [LINUX-390] Who's been reading our list...
. Joe Temple Distinguished Engineer Sr. Certified IT Specialist [EMAIL PROTECTED] 845-435-6301 295/6301 cell 914-706-5211 Home office 845-338-1448 Home 845-338-8794 Alan Cox [EMAIL PROTECTED] u.org.uk To Sent by: Linux on LINUX-390@VM.MARIST.EDU 390 Port cc [EMAIL PROTECTED] IST.EDU Subject Re: Fw: [LINUX-390] Who's been reading our list... 05/18/2006 01:17 PM Please respond to Linux on 390 Port [EMAIL PROTECTED] IST.EDU On Iau, 2006-05-18 at 09:51 -0400, Joseph Temple wrote: Yes tagging works, but you will find that the system z holds a lot more translations in a two tiered TLB and has tagging as well. Thus the System z does not have to retranslate as often. How many tags does the Z have in the TLBs ? -- For LINUX-390 subscribe / signoff / archive access instructions, send email to [EMAIL PROTECTED] with the message: INFO LINUX-390 or visit http://www.marist.edu/htbin/wlvindex?LINUX-390 -- For LINUX-390 subscribe / signoff / archive access instructions, send email to [EMAIL PROTECTED] with the message: INFO LINUX-390 or visit http://www.marist.edu/htbin/wlvindex?LINUX-390 attachment: pic05446.jpg attachment: pic18015.jpg
Re: Google out of capacity?
I agree.To google the computer that they have custom designed and programmed is their factor. It produces their product.Thus for them the development of this machine is like GM building their auto assembly line. Thought of in that light the extra development cost is a more reasonable expense than it would be for a business that uses computers as adjuncts to the actual manufacture of product. The point is that the customization has a direct rather than an indirect lever on their ability to produce. They get to a $1000 per server because they don't even buy blades, or rack optimized Intel servers, they buy parts. Of course they have to amortize their development expense and assembly cost which adds to people dollars. The development is amortized over a volume of thousands. Thus they have a superniche. Once they have their infrastructure in place for search it is much easier for them to build applications to work there than to customize new infrastructure around other applications. Finally they save on development because they use long standing research in parallel processing, which has been popular because of entry cost for at least 20 years. The questions become: Does the infrastructure break as some level of scale? If seems that some cracks are appearing. Are they too inefficient with environmentals? This changes with oil @ ~$75 per barrel. Are they using too many people? This one would be hard to extract. The people they use to run this thing are their factory workers which business tend to view differently than IT operations. In another solution at least some of the same people would likely have other tasks, but would not necessarily go away. Finally there is the question Can anything else do the job? The answer to this is not clear either. My guess is that a successful replacement would either be a standard yet still very distributed solution like blades, or a niche technology solution with lots of cores per chip, or a heavily virtualized system like z Linux. or some hybrid. Simply replacing their nodes with fewer larger nodes will probably not do it. Joe Temple Distinguished Engineer Sr. Certified IT Specialist [EMAIL PROTECTED] 845-435-6301 295/6301 cell 914-706-5211 Home office 845-338-1448 Home 845-338-8794 shogunx [EMAIL PROTECTED] ak.ath.cx To Sent by: Linux on LINUX-390@VM.MARIST.EDU 390 Port cc [EMAIL PROTECTED] IST.EDU Subject Re: Google out of capacity? 05/05/2006 02:02 AM Please respond to Linux on 390 Port [EMAIL PROTECTED] IST.EDU On Fri, 5 May 2006, Vic Cross wrote: On 05/05/2006, at 5:53am, Fargusson.Alan wrote: A long time ago I read that they did TCO studies, and found it less costly to buy lots of low cost hardware over buying fewer high cost systems. A long time ago is the point. When I read similar, the server count was around 8000 -- it would seem that they've grown considerably beyond that now. I doubt they've updated their TCO analysis accordingly... :) The real question is could a z9 outprocess existing clusters, out scale them at the same rate, and do so in such a fashion as to make it attractive to google to abandon its own in-house OS? Cheers, Vic Cross -- For LINUX-390 subscribe / signoff / archive access instructions, send email to [EMAIL PROTECTED] with the message: INFO LINUX-390 or visit http://www.marist.edu/htbin/wlvindex?LINUX-390 sleekfreak pirate broadcast http://sleekfreak.ath.cx:81/ -- For LINUX-390 subscribe / signoff / archive access instructions, send email to [EMAIL PROTECTED] with the message: INFO LINUX-390 or visit http://www.marist.edu/htbin/wlvindex?LINUX-390 -- For LINUX-390 subscribe / signoff / archive access instructions, send email to [EMAIL PROTECTED] with the message: INFO LINUX-390 or visit http://www.marist.edu/htbin/wlvindex?LINUX-390
Re: Google out of capacity?
Actually, in Google's case there is some assembly required which goes beyond unpacking the server and sticking it in a rack, unless they now have a supplier for their nodes. I had the impression that they assembled the nodes themselves. In any case the lead time may be longer leading to bigger inventory on hand, particularly if the throw away broken stuff rather than repair. Joe Temple Distinguished Engineer Sr. Certified IT Specialist [EMAIL PROTECTED] 845-435-6301 295/6301 cell 914-706-5211 Home office 845-338-1448 Home 845-338-8794 Tim Hare [EMAIL PROTECTED] te.fl.us To Sent by: Linux on LINUX-390@VM.MARIST.EDU 390 Port cc [EMAIL PROTECTED] IST.EDU Subject Re: Google out of capacity? 05/05/2006 08:40 AM Please respond to Linux on 390 Port [EMAIL PROTECTED] IST.EDU I am not sure I remember this right, but I think they said the pay less then $1,000 per server. They can buy a lot of $1,000 systems for the cost of one z9. I'd say that's true in one sense - without the TCO calculations. But in another sense - how much does it cost for a virtual Linux server? Assuming (I know, I know.. ) their z/VM box has some capacity left, it costs very little to create one more Linux image, and you don't have to wait for someone to bring one from the storage room / warehouse. Note that I'm assuming Google buys extra hardware in advance, installing it as needed, rather than waiting until they need it to buy it. If they wait until the last minute to buy it, then you have to figure in delivery / shipping costs. Tim Hare Senior Systems Programmer Florida Department of Transportation (850) 414-4209 -- For LINUX-390 subscribe / signoff / archive access instructions, send email to [EMAIL PROTECTED] with the message: INFO LINUX-390 or visit http://www.marist.edu/htbin/wlvindex?LINUX-390 -- For LINUX-390 subscribe / signoff / archive access instructions, send email to [EMAIL PROTECTED] with the message: INFO LINUX-390 or visit http://www.marist.edu/htbin/wlvindex?LINUX-390
Re: Google out of capacity?
Utilization of something as large as Google is an interesting issue. Given the structure that Samuel explained: they use a distributed processing model where a master server(s) sends jobs to any available node that has enough available CPU cycles. The ability to utilize all those processors will depend on how well the master can load balance the work on the nodes. This usually depends on the variability in the size and shape of the work to be done, the length (delay) of the feedback path that delivers the I have cycles info to the master and the affinities of various types of work to particular pools of capacity. Generally speaking high variability, long feed back path and affinity scheduling will lead to reduced utilization.Round robin routing will not have the feedback and affinity drivers, but will have some servers clogged while others are idle given enough workload variability. In general the more servers there are in the cluster, the stronger these effects and the lower the utilization. On the other hand if the master is able to break the work into relatively uniform packages and distribute them the utilization can be quite high. Depends on the load. Joe Temple Distinguished Engineer Sr. Certified IT Specialist [EMAIL PROTECTED] 845-435-6301 295/6301 cell 914-706-5211 Home office 845-338-1448 Home 845-338-8794 Little, Chris [EMAIL PROTECTED] hs.orgTo Sent by: Linux on LINUX-390@VM.MARIST.EDU 390 Port cc [EMAIL PROTECTED] IST.EDU Subject Re: Google out of capacity? 05/05/2006 01:46 PM Please respond to Linux on 390 Port [EMAIL PROTECTED] IST.EDU If the servers are running at a certain percentage of capacity, how would virtualization help? z or otherwise? -Original Message- From: Alan Altmark [mailto:[EMAIL PROTECTED] Sent: Friday, May 05, 2006 12:43 PM To: LINUX-390@VM.MARIST.EDU Subject: Re: Google out of capacity? On Friday, 05/05/2006 at 10:09 EST, Rich Smrcina [EMAIL PROTECTED] wrote: I think it's a safe bet that many 54-way z9's would be required, and lots of fully loaded DS8000's, with the new 4G Ficon (insert tool man growl here). It would be a sweet coup if there were any interest. Picture it: The year is 2050. The public's demand for information has continued to grow unabated, and there are 9+B people on the planet (source: US Census Bureau). The landfills are full of broken servers and the deserts are covered with solar collectors to fuel the server farms. These servers are located underground as there is no more space above ground. The heat from all the servers has altered the climate and raised the ocean levels. All of our homes are on stilts. Or, they could choose some form of virtualization and save us all. (He Who Must Not Be Annoyed says that z is not the answer for Google.) -- Chuckie -- For LINUX-390 subscribe / signoff / archive access instructions, send email to [EMAIL PROTECTED] with the message: INFO LINUX-390 or visit http://www.marist.edu/htbin/wlvindex?LINUX-390 -- For LINUX-390 subscribe / signoff / archive access instructions, send email to [EMAIL PROTECTED] with the message: INFO LINUX-390 or visit http://www.marist.edu/htbin/wlvindex?LINUX-390 -- For LINUX-390 subscribe / signoff / archive access instructions, send email to [EMAIL PROTECTED] with the message: INFO LINUX-390 or visit http://www.marist.edu/htbin/wlvindex?LINUX-390
Re: Google out of capacity?
IBM probably could build them, whether we could sell them at price Google could afford is another issue... Does anyone know how many of what class of servers are being used? Also, my guess is that some sort of hybrid might be the answer. That is some of the clusters may lend themselves to virtualization more than others, yielding variable leverage for different platforms. Joe Temple Distinguished Engineer Sr. Certified IT Specialist [EMAIL PROTECTED] 845-435-6301 295/6301 cell 914-706-5211 Home office 845-338-1448 Home 845-338-8794 Alan Cox [EMAIL PROTECTED] u.org.uk To Sent by: Linux on LINUX-390@VM.MARIST.EDU 390 Port cc [EMAIL PROTECTED] IST.EDU Subject Re: Google out of capacity? 05/04/2006 11:25 AM Please respond to Linux on 390 Port [EMAIL PROTECTED] IST.EDU On Iau, 2006-05-04 at 09:31 -0500, Dave Jones wrote: But today is special - the CEO has admitted that the grand distributed PC approach hasn't worked. http://www.webmasterworld.com/forum30/34147.htm Funny but that doesn't seem to be what the original referenced material is about. Huge machine crisis? Is there a zSeries salesman in the room? I doubt IBM could build enough zSeries boxes in a decade to even match the existing infrastructure 8) -- For LINUX-390 subscribe / signoff / archive access instructions, send email to [EMAIL PROTECTED] with the message: INFO LINUX-390 or visit http://www.marist.edu/htbin/wlvindex?LINUX-390 -- For LINUX-390 subscribe / signoff / archive access instructions, send email to [EMAIL PROTECTED] with the message: INFO LINUX-390 or visit http://www.marist.edu/htbin/wlvindex?LINUX-390
Re: New z9 models
Nope... Some of us are too old for that, but we did play Adventure on VM... maze of twisty turny little passages all different. No graphics, just keystrokes and imagination. Joe Temple Distinguished Engineer Sr. Certified IT Specialist [EMAIL PROTECTED] 845-435-6301 295/6301 cell 914-706-5211 Home office 845-338-1448 Home 845-338-8794 Alan Altmark/Endicott/ [EMAIL PROTECTED] To Sent by: Linux on LINUX-390@VM.MARIST.EDU 390 Port cc [EMAIL PROTECTED] IST.EDU Subject Re: New z9 models 04/28/2006 12:51 PM Please respond to Linux on 390 Port [EMAIL PROTECTED] IST.EDU On Friday, 04/28/2006 at 10:05 EST, Tom Duerbusch [EMAIL PROTECTED] wrote: I know we are all virtual...really... But in what universe can you have a 20 sided dice? The pursuit of a 20 sided dice, would make the pursuit of free beer look as easy as turning on the tap (for those of us that don't work at a brewery). Dude! You never played Dungeons Dragons? Two 20-sided dice (icosahedra) give you a number between 00 and 99 with an even probability distribution. Picture at http://en.wikipedia.org/wiki/Icosahedron. Octahedra and dodecahedra (12-sided) are used as well. We even occasionally used the more famous hexahedron. I thought all computer geeks played DD :-) Alan Altmark z/VM Development IBM Endicott -- For LINUX-390 subscribe / signoff / archive access instructions, send email to [EMAIL PROTECTED] with the message: INFO LINUX-390 or visit http://www.marist.edu/htbin/wlvindex?LINUX-390 -- For LINUX-390 subscribe / signoff / archive access instructions, send email to [EMAIL PROTECTED] with the message: INFO LINUX-390 or visit http://www.marist.edu/htbin/wlvindex?LINUX-390
Re: Business Week Article
Does the Intel Rack contain a switch like a blade center has? If so this is a classic example of Infrastructure Simplification and a lot of the savings would be in network infrastructure and admin costs. Joe Temple Executive Architect Sr. Certified IT Specialist [EMAIL PROTECTED] 845-435-6301 295/6301 cell 914-706-5211 Home office 845-338-1448 Home 845-338-8794 Meanor, Tim [EMAIL PROTECTED] ott.com To Sent by: Linux on LINUX-390@VM.MARIST.EDU 390 Port cc [EMAIL PROTECTED] IST.EDU Subject Re: Business Week Article 06/30/2005 12:12 PM Please respond to Linux on 390 Port Well, in all fairness, they moved 60 web sites to zLinux, but also consolidated 500 applications from 560 x86 servers to a rack of 70 x86 servers running VMware. That entire migration is what is expected to save $10 million, not just the move to zLinux. -Original Message- From: Linux on 390 Port [mailto:[EMAIL PROTECTED] On Behalf Of Neale Ferguson Sent: Wednesday, June 29, 2005 9:42 AM To: LINUX-390@VM.MARIST.EDU Subject: Business Week Article BusinessWeek article tells the story of First National Bank of Nebraska consolidating 30 (Sun) Unix servers onto one mainframe. The shift boosted hardware-utilization rates to about 70% - and Kucera expects to save $10 million over five years. 'It's revolutionary,' he says. 'It's really good stuff. It paid for itself in a year.' http://www.businessweek.com/magazine/content/05_25/b3938622.htm -- For LINUX-390 subscribe / signoff / archive access instructions, send email to [EMAIL PROTECTED] with the message: INFO LINUX-390 or visit http://www.marist.edu/htbin/wlvindex?LINUX-390 -- For LINUX-390 subscribe / signoff / archive access instructions, send email to [EMAIL PROTECTED] with the message: INFO LINUX-390 or visit http://www.marist.edu/htbin/wlvindex?LINUX-390 -- For LINUX-390 subscribe / signoff / archive access instructions, send email to [EMAIL PROTECTED] with the message: INFO LINUX-390 or visit http://www.marist.edu/htbin/wlvindex?LINUX-390
Re: Business Week Article
The values that Uriel uses is a reasonable starting point when you don't know either the utilization or details about the application. It is not precisely the same as we use for sizing at IBM, but there is more to this than our guess about the z. The basis we use would make the Sun machine have a different capacity than the Aix machine so we are talking about an extrapolation on an extrapolation. In any case data can make the answer go in either direction and the Uriels estimate is not extreme. Joe Temple Executive Architect Sr. Certified IT Specialist [EMAIL PROTECTED] 845-435-6301 295/6301 cell 914-706-5211 Home office 845-338-1448 Home 845-338-8794 Meanor, Tim [EMAIL PROTECTED] ott.com To Sent by: Linux on LINUX-390@VM.MARIST.EDU 390 Port cc [EMAIL PROTECTED] IST.EDU Subject Re: Business Week Article 06/30/2005 12:10 PM Please respond to Linux on 390 Port What sorts of tests did you run to arrive at this formula? I'm just curious because it seems like a bit much to compare a single IFL to a 4xCPU AIX or Sun box. -Original Message- From: Linux on 390 Port [mailto:[EMAIL PROTECTED] On Behalf Of Uriel Carrasquilla Sent: Thursday, June 30, 2005 11:15 AM To: LINUX-390@VM.MARIST.EDU Subject: Re: Business Week Article John: The formula we use is 4xCPU (sun 1.2 GHz or aix 1.0 GHz) is the same as one IFL under z890 running zLinux under LPAR. If zVM, we take away 10% power for the zVM overhead. This is a rough estimate and gets refined upon the real circumstances. Products licensing based on number of CPU's get penalized but we spend more for the H/W (IBM is happy). There are other benefits that are more important, such as saving in head counts that I rather not get into but you can figure it out. Regards, [EMAIL PROTECTED] NCCI Boca Raton, Florida 561.893.2415 greetings / avec mes meilleures salutations / Cordialmente mit freundlichen Grüßen / Med vänlig hälsning McKown, John [EMAIL PROTECTED]To: LINUX-390@VM.MARIST.EDU insctr.com cc: Sent by: Linux onSubject: Re: Business Week Article 390 Port [EMAIL PROTECTED] IST.EDU 06/30/2005 09:35 AM Please respond to Linux on 390 Port -Original Message- From: Linux on 390 Port [mailto:[EMAIL PROTECTED] On Behalf Of Pieter Harder Sent: Thursday, June 30, 2005 8:18 AM To: LINUX-390@VM.MARIST.EDU Subject: Re: Business Week Article Run Oracle on the z/890 as a super server, (We just eliminated z/VM and Linux on the zSeries because nobody knew what to do with them John, Just curious, on what are they planning to run Oracle then? zOS? You guys mus have gobs of money Best regards, Pieter Harder We don't have gobs of money. That scenario is what is being touted by certain managers. Personally, I don't buy it, but what do I know? They say that Oracle is licensed by the number of processors in the box. And that a license for z/OS on two processors (regardless of which processor it is - z800, z890, z990, all the same cost) costs the same as a license for Windows running on two processors. This seems silly to me, but I cannot refute it. Also, at one time, we did test Oracle on Linux under z/VM (back when we had those products). The Oracle DBAs (who ran the test themselves) came to the conclusion that a z800 single IFL and 1 Gb of memory did not perform as well as a 10Gb Sun system with 10 processors. Well, duh! -- John McKown Senior Systems Programmer UICI Insurance Center Information Technology This message (including any attachments) contains confidential information intended for a specific individual and purpose, and its' content is protected by law. If
Re: Business Week Article
This comparison is also in the range of values that have been measured. However it is near the low end of the range for commercial work. We usually see this type of comparison when there are long pathlengths of work per byte processed, query oriented workloads with little locking or writes, or data base loads that exploit private caches and don't benefit from zSeries shared cache. A relatively small cache working set and high sustained utilization of the Intel processor will drive this kind of result. It can also occur when comparing a fully utilized dedicated intel machine to a IFL in an LPAR with a lot of other work going on. It is not a center value to use when you don't understand the application behavior or production utlization. Joe Temple Executive Architect Sr. Certified IT Specialist [EMAIL PROTECTED] 845-435-6301 295/6301 cell 914-706-5211 Home office 845-338-1448 Home 845-338-8794 Little, Chris [EMAIL PROTECTED] hs.orgTo Sent by: Linux on LINUX-390@VM.MARIST.EDU 390 Port cc [EMAIL PROTECTED] IST.EDU Subject Re: Business Week Article 06/30/2005 01:16 PM Please respond to Linux on 390 Port Were those Solaris/AIX servers underutilized? By a lot? Certainly workloads don't compare across platforms. I agree with that, but for database workloads I was told (informally) to assume a single z/900 IFL at a Pentium III 700mhz. Within reason, it seems correct. We are probably getting better than that, but certainly nowhere near what you are seeing. Low CPU/high IO (webserving, maybe?) might translate better to linux on zseries, I don't know. -Original Message- From: Uriel Carrasquilla [mailto:[EMAIL PROTECTED] Sent: Thursday, June 30, 2005 12:12 PM To: LINUX-390@VM.MARIST.EDU Subject: Re: Business Week Article Crhis: I suspect not all workloads are the same. What we are finding out is that loads are taking place much faster. That was a large pecentage of our usage before. We don't have any users signed on and we only run the zLinux image to service requests. It works for us and no one is complaining. Regards, [EMAIL PROTECTED] NCCI Boca Raton, Florida 561.893.2415 greetings / avec mes meilleures salutations / Cordialmente mit freundlichen Grüßen / Med vänlig hälsning Little, Chris [EMAIL PROTECTED]To: LINUX-390@VM.MARIST.EDU hs.org cc: Sent by: Linux onSubject: Re: Business Week Article 390 Port [EMAIL PROTECTED] IST.EDU 06/30/2005 11:48 AM Please respond to Linux on 390 Port Are you getting that much performance out of an IFL? Four times the performance of a single RISC unix server? I say that because the Oracle database we moved from an HP server with 2 PA-RISC 8600 (440mhz) cpus consumed a little more than one IFL(100%-120% on a two IFL LPAR). This was about in line with what IBM recommended. -Original Message- From: Uriel Carrasquilla [mailto:[EMAIL PROTECTED] Sent: Thursday, June 30, 2005 10:15 AM To: LINUX-390@VM.MARIST.EDU Subject: Re: Business Week Article John: The formula we use is 4xCPU (sun 1.2 GHz or aix 1.0 GHz) is the same as one IFL under z890 running zLinux under LPAR. If zVM, we take away 10% power for the zVM overhead. This is a rough estimate and gets refined upon the real circumstances. Products licensing based on number of CPU's get penalized but we spend more for the H/W (IBM is happy). There are other benefits that are more important, such as saving in head counts that I rather not get into but you can figure it out. Regards, [EMAIL PROTECTED] NCCI Boca Raton, Florida 561.893.2415 greetings / avec mes meilleures salutations / Cordialmente mit freundlichen Grüßen /
Re: Business Week Article
Part of the difference is the comparison of the HP to the Intel Box. If you take the z900 uniprocessor IFL to be 256 MIPS my first guess for the HP 440 MHz 2way would have been 290 MIPS, but for the PIII 700 MHz I would have come up with 83.5 MIPS. I do these comparisons by using extrapolation data we purchase from Ideas International or using Internal IBM extrapolations (they yield similar results but cover different machines), and then apply a middle of the road conversion factor and a 70% utlization factor. Any way the 290 MIPS would chew up an IFL and then some. I can't share the conversion factors with you because of agreements IBM has with benchmark councils and with Ideas. In any case I think your estimator is over valuing the PIII 700 MHz engine. Joe Temple Executive Architect Sr. Certified IT Specialist [EMAIL PROTECTED] 845-435-6301 295/6301 cell 914-706-5211 Home office 845-338-1448 Home 845-338-8794 Little, Chris [EMAIL PROTECTED] hs.orgTo Sent by: Linux on LINUX-390@VM.MARIST.EDU 390 Port cc [EMAIL PROTECTED] IST.EDU Subject Re: Business Week Article 06/30/2005 01:23 PM Please respond to Linux on 390 Port I would guess that the SQL is not very optimized. Also, the app pushes a lot of math that could be done client side to the database. 700 may be a somewhat low, but the HPUX server had two 440mhz PA-RISC 8?00 cpu's. When we moved it, it ate all of one IFL, plus a bit. It's bloated since then and has been chewing on a good portion of both recently. -Original Message- From: Rich Smrcina [mailto:[EMAIL PROTECTED] Sent: Thursday, June 30, 2005 12:15 PM To: LINUX-390@VM.MARIST.EDU Subject: Re: Business Week Article 700Mhz? That seems pretty low. Little, Chris wrote: Were those Solaris/AIX servers underutilized? By a lot? Certainly workloads don't compare across platforms. I agree with that, but for database workloads I was told (informally) to assume a single z/900 IFL at a Pentium III 700mhz. Within reason, it seems correct. We are probably getting better than that, but certainly nowhere near what you are seeing. Low CPU/high IO (webserving, maybe?) might translate better to linux on zseries, I don't know. -Original Message- From: Uriel Carrasquilla [mailto:[EMAIL PROTECTED] Sent: Thursday, June 30, 2005 12:12 PM To: LINUX-390@VM.MARIST.EDU Subject: Re: Business Week Article Crhis: I suspect not all workloads are the same. What we are finding out is that loads are taking place much faster. That was a large pecentage of our usage before. We don't have any users signed on and we only run the zLinux image to service requests. It works for us and no one is complaining. Regards, [EMAIL PROTECTED] NCCI Boca Raton, Florida 561.893.2415 greetings / avec mes meilleures salutations / Cordialmente mit freundlichen Grüßen / Med vänlig hälsning Little, Chris [EMAIL PROTECTED]To: LINUX-390@VM.MARIST.EDU hs.org cc: Sent by: Linux onSubject: Re: Business Week Article 390 Port [EMAIL PROTECTED] IST.EDU 06/30/2005 11:48 AM Please respond to Linux on 390 Port Are you getting that much performance out of an IFL? Four times the performance of a single RISC unix server? I say that because the Oracle database we moved from an HP server with 2 PA-RISC 8600 (440mhz) cpus consumed a little more than one IFL(100%-120% on a two IFL LPAR). This was about in line with what IBM recommended. -Original Message- From: Uriel Carrasquilla [mailto:[EMAIL PROTECTED] Sent: Thursday, June 30, 2005 10:15 AM To: LINUX-390@VM.MARIST.EDU Subject: Re: Business Week
Re: Business Week Article
This one is often either overlooked or overblown. zSeries zealots will talk about how the IO processors do the I/O for z and I/O being done in the cpus for others. While neither machine really DOES I/O in either the CPU or SAP (IOP), there is high priority set up code which must be done to start and finish I/O's. In the zSeries this code is done in the SAPs. The real benefit of doing this is that the code is always run on an engine which is running at low utilization reducing the wait time significantly. Also the I/O processing context does not muck up the L1 cache of the CPs. To avoid excessive wait time and cache working set problems it is sometimes (but not always) necessary to run UNIX and Intel machines at lower utilization. When this happens there is a clear advantage to the zSeries. The other side of the coin is that the zSeries cannot use the SAP cycles for CPU work which for any given infrastructure reduces the number of CPUs in the machine. For this reason the SAPs do not count toward the machines capacity or pricing beyond their inclusion in the infrastructure of the machine. Joe Temple Executive Architect Sr. Certified IT Specialist [EMAIL PROTECTED] 845-435-6301 295/6301 cell 914-706-5211 Home office 845-338-1448 Home 845-338-8794 Uriel Carrasquilla Uriel_Carrasquil To [EMAIL PROTECTED] LINUX-390@VM.MARIST.EDU Sent by: Linux on cc 390 Port [EMAIL PROTECTED] Subject IST.EDU Re: Business Week Article 06/30/2005 01:33 PM Please respond to Linux on 390 Port Chris: The z890 uses a separate CPU for all of the I/O's. Something we don't have on the aix/sun platform. The application does lots of I/O to get the data and the server had tons of cache which was of little value for our purposes. We trimmed memory to just what was needed, mostly PGA plus 128 MB for the OS plus a bonus 128 MB to feel good. Now, instead of CPU cycles on my main processor managing large cache, the work is shifted to the I/O CPU. We run aix/sun at about 60% to keep customers happy. Now I can run busier and we meet our requirements. We are also running Apache, SMB and NFS. We collect a lot of data on a daily basis and load into our dbms. Regards, [EMAIL PROTECTED] NCCI Boca Raton, Florida 561.893.2415 greetings / avec mes meilleures salutations / Cordialmente mit freundlichen Grüßen / Med vänlig hälsning Little, Chris [EMAIL PROTECTED]To: LINUX-390@VM.MARIST.EDU hs.org cc: Sent by: Linux onSubject: Re: Business Week Article 390 Port [EMAIL PROTECTED] IST.EDU 06/30/2005 01:16 PM Please respond to Linux on 390 Port Were those Solaris/AIX servers underutilized? By a lot? Certainly workloads don't compare across platforms. I agree with that, but for database workloads I was told (informally) to assume a single z/900 IFL at a Pentium III 700mhz. Within reason, it seems correct. We are probably getting better than that, but certainly nowhere near what you are seeing. Low CPU/high IO (webserving, maybe?) might translate better to linux on zseries, I don't know. -Original Message- From: Uriel Carrasquilla [mailto:[EMAIL PROTECTED] Sent: Thursday, June 30, 2005 12:12 PM To: LINUX-390@VM.MARIST.EDU Subject: Re: Business Week Article Crhis: I suspect not all workloads are the same. What we are finding out is that loads are taking place much faster. That was a large pecentage of our usage before. We don't have any users signed on and we only run the zLinux image to service requests. It works for us and no one is complaining. Regards, [EMAIL PROTECTED] NCCI Boca Raton,
Re: CPU Comparison
Larry, the conversion rate for Intel platforms to zSeries ranges from 3 MHz per MIPS to 30 MHz per MIPS depending on how much data you are pushing through the caches and out much serialization (context switches, locks, etc.). This is typically lower for the web servers and higher for the backend. Because zSeries capacity is almost always a mixed workload capacity the utilization peak and utilization variation on an interval by interval basis comes into play. This is because you have to configure each blade for its peak, but you configure zSeries for the composite peak which is almost always less or can be managed to be much less than the sum of the peaks. Because some workloads tend to saturate blades at lower utilization than zSeries the composite utilization peak is usually quite low. Joe Temple Executive Architect Sr. Certified IT Specialist [EMAIL PROTECTED] 845-435-6301 295/6301 cell 914-706-5211 Home office 845-338-1448 Home 845-338-8794 Davis, Larry [EMAIL PROTECTED] ielsenMedia.com To Sent by: Linux on LINUX-390@VM.MARIST.EDU 390 Port cc [EMAIL PROTECTED] IST.EDU Subject CPU Comparison 03/18/2005 11:42 AM Please respond to Linux on 390 Port I know that this has been asked in many different ways, but has anyone actually put together a comparison chart or list of CPU speeds for zSeries, I32, I64, and other processors. I have a department that wants to have a group of dedicated Blades assigned to specific clients. We want to show a cost to Performance ratio for purchasing blades and scaling a zSeries system. The application uses Java, in the backend to run the queries and retrieve the data, and Web servers to interface to a client and present the results. Larry Davis, Senior Systems Engineer Nielsen Media Research 813-366-2380 -- For LINUX-390 subscribe / signoff / archive access instructions, send email to [EMAIL PROTECTED] with the message: INFO LINUX-390 or visit http://www.marist.edu/htbin/wlvindex?LINUX-390 -- For LINUX-390 subscribe / signoff / archive access instructions, send email to [EMAIL PROTECTED] with the message: INFO LINUX-390 or visit http://www.marist.edu/htbin/wlvindex?LINUX-390
Re: Why Zseries
Ralph, How often do you change releases? If you go to Microsoft do your programmers understand that you are going into a keep up or die culture? Yes the application programming fashionable, and hardware is cheap, but microsoft does not provide long term support with regard to release levels. The shift can be costly either in revamping how you do everything, or in finding yourself unsupported because microsoft moves faster that you are able to keep up. Joe Temple Executive Architect Sr. Certified IT Specialist [EMAIL PROTECTED] 845-435-6301 295/6301 cell 914-706-5211 Home office 845-338-1448 Home 845-338-8794 -- For LINUX-390 subscribe / signoff / archive access instructions, send email to [EMAIL PROTECTED] with the message: INFO LINUX-390 or visit http://www.marist.edu/htbin/wlvindex?LINUX-390
Re: Why Zseries
Let me add to what Joe added. When you combine the low utilization that many (not all) dense rack mounted servers run at it becomes even easier for z to win the througput/KVA race. Even if we don't include non production servers and look at clusters for a single application, the peak composite utilization of distributed solutions is very often under 25%. Adding non production servers, mixing applications and doing workload management with VM can increase utilization leverage on relative capacity well above the 4 to 1 indicated by the production data. Joe Temple Executive Architect Sr. Certified IT Specialist [EMAIL PROTECTED] 845-435-6301 295/6301 cell 914-706-5211 Home office 845-338-1448 Home 845-338-8794 Joe Poole [EMAIL PROTECTED] omTo Sent by: Linux on LINUX-390@VM.MARIST.EDU 390 Port cc [EMAIL PROTECTED] IST.EDU Subject Re: Why Zseries 02/10/2005 01:46 PM Please respond to Linux on 390 Port On Thursday 10 February 2005 01:32 pm, Levy, Alan wrote: David - thanks. This is what that I was looking for. Let me add to what Dr. David said with a few metrics. The densely packed servers of today draw between .1 and .7 kVA of power. If you take .3 kVA as an average, it takes only 15 servers to match the 4.5 kVA of a z900. Add more CPUs and internal disk to the servers, and it's not uncommon to find a single server rack that exceeds the power requirements of the zSeries with a Shark thrown in for good measure. I've heard that it's become difficult to add the required power to some of the buildings in NYC to support the server induced power load. -- For LINUX-390 subscribe / signoff / archive access instructions, send email to [EMAIL PROTECTED] with the message: INFO LINUX-390 or visit http://www.marist.edu/htbin/wlvindex?LINUX-390 -- For LINUX-390 subscribe / signoff / archive access instructions, send email to [EMAIL PROTECTED] with the message: INFO LINUX-390 or visit http://www.marist.edu/htbin/wlvindex?LINUX-390
Re: Why Zseries
Speaking of KVA, has anyone else heard about anyone hooking up the old machine floor plumbing to chillers in order get cool enough air on the floor to cool dense racks or blade centers. Just wondering if what I heard is a rumour, at fact or a mainframe geek joke :-) Joe Temple Executive Architect Sr. Certified IT Specialist [EMAIL PROTECTED] 845-435-6301 295/6301 cell 914-706-5211 Home office 845-338-1448 Home 845-338-8794 Joe Poole [EMAIL PROTECTED] omTo Sent by: Linux on LINUX-390@VM.MARIST.EDU 390 Port cc [EMAIL PROTECTED] IST.EDU Subject Re: Why Zseries 02/10/2005 01:46 PM Please respond to Linux on 390 Port On Thursday 10 February 2005 01:32 pm, Levy, Alan wrote: David - thanks. This is what that I was looking for. Let me add to what Dr. David said with a few metrics. The densely packed servers of today draw between .1 and .7 kVA of power. If you take .3 kVA as an average, it takes only 15 servers to match the 4.5 kVA of a z900. Add more CPUs and internal disk to the servers, and it's not uncommon to find a single server rack that exceeds the power requirements of the zSeries with a Shark thrown in for good measure. I've heard that it's become difficult to add the required power to some of the buildings in NYC to support the server induced power load. -- For LINUX-390 subscribe / signoff / archive access instructions, send email to [EMAIL PROTECTED] with the message: INFO LINUX-390 or visit http://www.marist.edu/htbin/wlvindex?LINUX-390 -- For LINUX-390 subscribe / signoff / archive access instructions, send email to [EMAIL PROTECTED] with the message: INFO LINUX-390 or visit http://www.marist.edu/htbin/wlvindex?LINUX-390
Re: Why Zseries
Kielek is correct, but consider this. 1. Given the availability of the application, there is a small difference between Linux on z and Linux on Intel simply because the zSeries reliability takes the hardware multiplier on availability closer to 1. 2. Yes we can configure the Intel with redundant servers to mitigate the reliability difference. a. In the case of stateful applications such failovers are small outages because they take measurable time (up to minutes or even hours on thorny situations). In this case the reduction of the number outages by better hardware availability still helps. b. In the case of stateless applications you need less redundant hardware on z than on intel because you can effectively run the z at higher utilization than the intel machine for many applications. This is because many workloads cause the Intel boxes to saturate a relatively low utilization. When this happens it takes more than n+1 boxes to deliver n+1 availability is than one unless the Intel machine is run at very low utilization. Since the zSeries solution is more like to be CPU bound the utilization at which n+1 can be n+1 is higher. c.) Since the failing component is most likely the application or the linux, there is the opportunity to set the Linux on z farm up in such a way that the remaining linux images get the capacity that the failed linux had. In other words a redundant linux/application instance is provided but still less redundant capacity is required. This depends on being able to detect the failure and restart the failed linux with reduced resources until it is ready to accept the load. That is the failed linux gets hard capped and the remaining soft capped images grab the resulting whitespace. Joe Temple Executive Architect Sr. Certified IT Specialist [EMAIL PROTECTED] 845-435-6301 295/6301 cell 914-706-5211 Home office 845-338-1448 Home 845-338-8794 Kielek, Samuel [EMAIL PROTECTED] rriott.comTo Sent by: Linux on LINUX-390@VM.MARIST.EDU 390 Port cc [EMAIL PROTECTED] IST.EDU Subject Re: Why Zseries 02/10/2005 01:46 PM Please respond to Linux on 390 Port It is important to also understand that Linux is not capable (at least today) of directly exploiting many of those hardware benefits, especially in terms of the mainframes RAS features. That is to say, there can be instances where the mainframe is up and chugging along, VM is doing just fine, but Linux is dead in the water for example. Unfortunately the original question is simply too vague. Comparing Linux on zSeries to x86, pSeries, SPARC, etc. (if that was question) is as much a philosophical exercise as it is technical. What really needs to be focused on is the task that you wish to run, how it works and what it needs to perform its job(s). Then we can evaluate the pros and cons to running that task on each of the OS/architecture combinations. -Sam -Original Message- From: Linux on 390 Port [mailto:[EMAIL PROTECTED] On Behalf Of Adam Thornton Sent: Thursday, February 10, 2005 12:39 PM To: LINUX-390@VM.MARIST.EDU Subject: Re: Why Zseries On Feb 10, 2005, at 10:58 AM, Levy, Alan wrote: Actually, I probably phrased my question wrong. What I was looking for is why choose Zseries Linux over ANY OTHER operating system's Linux ? We're getting closer, but: zSeries is an architecture, not an operating system, and Linux doesn't generally run *on* an operating system[0], it *is* an operating system. That said, this looks like Question 2, if you substitute architecture's for operating system's. In which case: hardware reliability and fault tolerance, excellent sustained I/O capacity, potential for in-the-box really-high-speed interconnect with (usually z/OS) big databases in another LPAR, humongous consolidation opportunities if you have lightly-loaded guests, and incredibly rapid deployment of machines for testing and development. Adam [0] Unless run in a virtualization environment, like z/VM or VMware or Virtual PC, or Hercules, which is an emulation environment running under some OS. -- For LINUX-390 subscribe / signoff / archive access instructions, send email to [EMAIL PROTECTED] with the message: INFO LINUX-390 or visit http://www.marist.edu/htbin/wlvindex?LINUX-390 -- For LINUX-390 subscribe / signoff / archive access instructions, send email to [EMAIL PROTECTED] with the message: INFO LINUX-390 or visit http://www.marist.edu/htbin/wlvindex?LINUX-390
Re: need to compare apples to oranges (HP Unix CPU to zVM IFL CPU with Oracle)
Ken, Richard is on the right track here. Short of a Size390 sizing I can tell you that the range of relative capacity between a z900 is quite broad. The actual result will depend on what kind of workload is being done (query only, some updates, heavy transactional), the cache working set size (how much stress the workload puts on the hardware caches), and the usage(what is the peak utilization and does the cpu utilization for the various DB's peak simultaneously?). Size390 will send you a questionaire which will allow Techline to assess the workload and utilization conversion factors for your case.Your IBM rep should be able to put you in touch with a FTSS (field techie) who can either put you in touch with Techline or run a quicksizer tool to give you a first guess. If you don't have an IBM rep email [EMAIL PROTECTED] and let me know where you are located. Joe Temple Sr. Certified IT Specialist [EMAIL PROTECTED] 845-435-6301 295/6301 cell 914-706-5211 home 845-338-8794 richard truett [EMAIL PROTECTED]To: [EMAIL PROTECTED] .netcc: Sent by: Linux onSubject: Re: need to compare apples to oranges (HP Unix CPU to zVM IFL CPU with 390 Port Oracle) [EMAIL PROTECTED] IST.EDU 07/13/2004 11:21 AM Please respond to Linux on 390 Port Ken, IBM has a couple of modleing tools available to size the workload on zLinux and IFLs/Memory. If you have a locak IBM rep or business partner rep You may want to ask about the Size390 or New Workload sizing that IBM Techline performs. This service is no cost and can assist in getting an estimate to the z900 requirements. If you do not have these contacts, let me know and we can discuss off-list. ---Original Message--- From: Linux on 390 Port Date: 07/13/04 10:14:48 To: [EMAIL PROTECTED] Subject: need to compare apples to oranges (HP Unix CPU to zVM IFL CPU with Oracle) Hi, We are looking at a pilot project to test an Oracle database running on Linux/zVM. Currently we have about five applications that run on various HP Unix servers. Each of these applications connect to their own Oracle instance. Each instance is about 300GB, so we have 1.5TB for the databases. Our test would be to move the five Oracle instances to a Linux/zVM server running on an IFL on a z900. We will have one 300GB copy of the database, and each of the five instances would appear to have their own copy since the updates for each instance will be intercepted and written to a private area. The Unix group says that the Oracle instances consume between two and three CPUs on a HP Superdome 750Mhz box. The Project Office wants to know how that consumption would compare to a z900 IFL. We said that we really need to perform the pilot to get the numbers, but they said they really need the numbers before we can do the pilot. Does anyone know how to compare the CPUs between the two platforms? The Project people are also worried that the VM overhead will result in slow response times. We can try and perform a test via a standard script on each box. Has experience in the performance to be gained/lost between the two platforms? On the Oracle side, if I had a database of 300GB with an instance name of ORA1, and I want to change the instance name to ORA2, how many records need to be changed? If the instance name is connected to every record, then this project will have trouble since we are trying to share the 300GB base with multiple instances. We would use the I/O intercept software to write the changes to a private area, and if it needs to update 300GB of data, then it is not a feasible solution. The DBA group is dubious of this concept (read project), so we need to demonstrate that it will work. We do not expect to save money in the hardware costs of this project. The saving should come from the flexibility to deploy new images, and the quicker turnaround for the database restores with the I/O intercept software. However, if we are going to need six IFLs to run this DB, then it is unlikely they will let us proceed no matter how much flexibility we could gain. Any information would be appreciated. Thanks, Ken Vance Amadeus -- For LINUX-390 subscribe / signoff / archive access instructions, send email to [EMAIL PROTECTED] with the message: INFO LINUX-390 or visit http://www.marist.edu/htbin/wlvindex?LINUX-390 -- For LINUX-390 subscribe / signoff / archive access instructions, send email to [EMAIL PROTECTED] with the message: INFO LINUX-390 or visit
Re: Performance with Multiple CPUs
I think the answer is that zSeries Linux scales to more than 4 processors. This is obviously workload dependent. Not sure what you mean by negative if you mean that 4 processors get less work done than 3, probably won't see that on zSeries. If you mean that you get less than 4X the uniprocessor performance you will see that on most workloads on virtually all machines. If you mean that you see less throughput on linux 4ways than on windows 4 ways, you may or may not see it on zLinux. This will depend on whether the z hardware overcomes whatever software advantage that Windows has. (The two effects will mitigate each other but are not really related, so there is no single answer.) Joe Temple Sr. Consulting Sr. Certified IT Specialist [EMAIL PROTECTED] 845-435-6301 295/6301 cell 914-706-5211 home 845-338-8794 Walter Wojcik [EMAIL PROTECTED]To: [EMAIL PROTECTED] com cc: Sent by: Linux onSubject: Performance with Multiple CPUs 390 Port [EMAIL PROTECTED] IST.EDU 06/22/2004 11:58 AM Please respond to Linux on 390 Port Does anyone have experience running zLinux in an LPAR which has multiple CPUs defined? We have experienced negative performance characteristics with Intel Linux versions (RedHat, SUSSE) when the Intel machine had 4 processors. We were wondering if the same performance degradations appear on the mainframe. Walter E. Wojcik [EMAIL PROTECTED] phone: (781) 301-2000 fax: (781) 301-2001 -- For LINUX-390 subscribe / signoff / archive access instructions, send email to [EMAIL PROTECTED] with the message: INFO LINUX-390 or visit http://www.marist.edu/htbin/wlvindex?LINUX-390 -- For LINUX-390 subscribe / signoff / archive access instructions, send email to [EMAIL PROTECTED] with the message: INFO LINUX-390 or visit http://www.marist.edu/htbin/wlvindex?LINUX-390
Re: Second Wind for Big Iron
Actually, Blue Gene is neither a new sort of machine nor a mainframe. It is essentially a very large Scalable Parallel (from around 1992) machine using Linux and Intel instead of AIX and Power.Thus it is a new sort of SP.I think using it as the example of big iron causes the article to miss the point. Linux on z, USS, Java on z, and z with tcp/ip networking are the types of things that make the new mainframe. System structure, RAS and LPARs on p690 are much better non z examples than Blue Gene. They did get the size right though; Blue Gene is Big Iron. Joe Temple Sr. Certified IT Specialist [EMAIL PROTECTED] 845-435-6301 295/6301 cell 914-706-5211 home 845-338-8794 Phil Payne [EMAIL PROTECTED]To: [EMAIL PROTECTED] arch.comcc: Sent by: Linux onSubject: Re: Second Wind for Big Iron 390 Port [EMAIL PROTECTED] IST.EDU 03/23/2004 04:02 AM Please respond to Linux on 390 Port http://www.businessweek.com/magazine/content/04_13/b3876068.htm Absolute twaddle. System/360 was the world's first open system. Principles of Operation and the Channel OEMI manuals permitted plug compatibility - OS/360 was public domain and the source code was freely available from IBM for the cost of the tapes. Open source was, if you like, the reason for System/360's initial success. And, IMO, OCO is one of the reasons for its decline. And the article clearly implies that Blue Gene is the new sort of mainframe, with the old sort headed for extinction. Hmm. Not according to what I'm hearing about IBM's plans. -- Phil Payne http://www.isham-research.com +44 7785 302 803 -- For LINUX-390 subscribe / signoff / archive access instructions, send email to [EMAIL PROTECTED] with the message: INFO LINUX-390 or visit http://www.marist.edu/htbin/wlvindex?LINUX-390 -- For LINUX-390 subscribe / signoff / archive access instructions, send email to [EMAIL PROTECTED] with the message: INFO LINUX-390 or visit http://www.marist.edu/htbin/wlvindex?LINUX-390
Re: Perpetuating Myths about the zSeries
The best way to understand is to take measurements of the running production systems. There are many tools for doing this and you may already be gathering at least the data that you would need. The way to look at utilization is to plot the utilization on intervals for a peak period, day or week depending on the data you get. You should be looking for intervals around 15 minutes in length. (Over 30 minutes smooths things too much and under 5 minutes tends to visually (and mathematically) hide the troughs. This is because we use the peak to do our sizings and as intervals get shorter the peak approaches 100% regardless of the utilization. (utilization is a statistic, at the cycle level the machine is either busy or its not. For short enough intervals the utilization data will either be zero or 100%.) Anyway, you then stack these graphs in a stacked bar or area chart and then note the composite peak. This will tell you what your aggregate utilization is. If all the servers peak at the same time, then the peak utilization on each one has to be low to get favorable consolidation effects. The second thing to do is to look at the saturation curve for the servers in question. Gather throughput data on the same intervals as your utilization data. (This can be network data or packet rates if you don't have anything else). Plot the throughput v utilization for each server. You are looking to see if the curve bends over or saturates at higher utilization. I like to plot linear, power and logaritmic trends through the data. Usually the power curve has the best fit, but the linear and logarithmic curves provide bounds with which to compare it. The more linear the data is the more cpu intense the application is, and therefore the lower the utilization has to be to get a good conversion ratio. If the curve is bent over, the average utilzation is low, and the workload peaks at an off time you have an ideal candidate. You can also start gathering I/O and context switch rates. High rates here usually indicate non cpu intense or mixed applications. IBM has people who can help you get the data and analyze it. Joe Temple [EMAIL PROTECTED] 845-435-6301 295/6301 cell 914-706-5211 home 845-338-8794 Eric Sammons [EMAIL PROTECTED]To: [EMAIL PROTECTED] t.frb.org cc: Sent by: Linux onSubject: Re: Perpetuating Myths about the zSeries 390 Port [EMAIL PROTECTED] IST.EDU 10/31/2003 07:55 AM Please respond to Linux on 390 Port What about memory intensive? And how do you gage the CPU intensive applications? For example we are planning to migrate some of our Solaris (SPARC) applications off of SPARC and into the z/VM Linux world. If I am looking at candidates for this migration I see systems (SPARC) with 10 - 30 percent utilization. What happens when I decide these word loads are good candidates with their low cpu usage on the SPARC platform but then install them into the Z environment and find out that they now have a cpu usage of 80 - 90 percent? Is this possible? Is there a good way to judge what applications on a given platform might be best suited for migration? Right now I am recommending that any candidate first do a QA of their application in the Z environment prior to do doing the full and final migration. thanks! Eric Sammons (804)697-3925 FRIT - Infrastructure Engineering Post, Mark K [EMAIL PROTECTED] Sent by: Linux on 390 Port [EMAIL PROTECTED] 10/30/2003 04:49 PM Please respond to Linux on 390 Port To: [EMAIL PROTECTED] cc: Subject:Re: Perpetuating Myths about the zSeries My answer was, and still is (and likely always will be) avoid any application that is CPU intensive. Yes, the zSeries has gotten faster, but so has Intel. The price-performance curve for CPU intensive work still favors Intel. I've seen nothing in the IBM announcements that would lead me to change any of the recommendations I've been making for the last 3 years. Unless and until the price-performance curve for zSeries matches that of Intel (or comes a couple of orders of magnitude closer), I will continue to make the same recommendations. Mark Post -Original Message- From: Jim Sibley [mailto:[EMAIL PROTECTED] Sent: Wednesday, October 29, 2003 7:31 PM To: [EMAIL PROTECTED] Subject: Re: Perpetuating Myths about the zSeries -snip- Linux on all sorts of platforms was just a gleam in someone's eye 5 years ago. It started getting pushed on the zSeries 3 years ago and the software and hardware have made great strides in the last 3 years. So CGI may not be appropriate today. So what is there we said was not appropriate 2 or 3 years ago that may be appropriate today on Linux zSeries?
Re: Perpetuating Myths about the zSeries
I don't know of any plans to make RMF-PM available on other platforms. I will look around, but it will be a week or so; others may be able to help sooner. Joe Temple [EMAIL PROTECTED] 845-435-6301 295/6301 cell 914-706-5211 home 845-338-8794 David Boyes [EMAIL PROTECTED]To: [EMAIL PROTECTED] e.net cc: Sent by: Linux onSubject: Re: Perpetuating Myths about the zSeries 390 Port [EMAIL PROTECTED] IST.EDU 10/31/2003 11:48 AM Please respond to Linux on 390 Port What about memory intensive? And how do you gage the CPU intensive applications? For example we are planning to migrate some of our Solaris (SPARC) applications off of SPARC and into the z/VM Linux world. Something that occurred to me (and since Joe Temple is kindly answering questions): are there any plans to make the RMF-PM data collection agent available on platforms other than zLinux? While it's not the best tool available, it'd be handy to be able to do before/after comparisons with the same tools reporting to FCON/Perf Toolkit. Right now I am recommending that any candidate first do a QA of their application in the Z environment prior to do doing the full and final migration. Never a bad idea, particularly given your employer...8-) -- db
Re: Perpetuating Myths about the zSeries
Actually, while I/O is the classic example of why processor speed is not everything, you don't have to move that far beyond the processor itself to show this. Note that the various types of servers have different size and structures of L1, L2, L3, caches and memory interfaces. Also note that the memory latency and bandwidth varies from machine to machine. Finally note that the faster the processor the higher the latencies are in terms of number of cycles. Benchmarks are sensitive to this, but not uniformly. For example TPC-C got a big boost from 8MB L2 caches, but SPECint is almost totally insensitive to L2 size. Real work has a tendency to show even more variability. That is there are more daemons,etc. chewing up cache space causing more misses, and the code is not as extensively tuned. This tide will float all boats (even if you run tpc-c for a living don't expect the tuned rates if you also are running security, monitoring, accounting,etc.). However, the differences in memory hierarchy will cause some machines to be impacted more than others. There is really no way except experience to tell how an application treats the memory hierarchy in this regard. As a result defining relative capacity with a single metric is not possible, and any benchmark or metric that is suggested for this will not match the real world, which will exhibit more dynamic variability both with time and workload. On thing is for sure though: If a cache is blown, during the miss time the processor is 100% busy as measured by normal means, and the throughput is zero. You can see if this is happening to an significant extent by plotting throughput v processor utilization. (Throughput can come from the application or be estimated by network data rate). Draw linear, power and logaritmic trends through the data. If the best trend is linear and the line intercepts the vertical axis near or below the origin then there is little or no saturation and little pressure on the memory hierarchy. If the best trend is logarithmic then there is heavy saturation and chances are that the workload is blowing the caches as load is applied. If the power curve fits the best the answer is somewhere in the middle. As the exponent of the power curve approaches 1 the workload is exhibiting less saturation because the trend becomes linear. The more saturation exhibited the higher the utilization at which zLinux consolidation is viable. This is because the raw conversion factor is typically better for more saturated workloads. My point is there is no such thing as a metric which defines the relative capacity of various servers because of the differences in their memory hierarchies. Processor speed is an indicator, but it is insufficient to do any real comparisons. The processor speeds are what they are. You can measure them with MHz, BOGOMIPS, SPECint, or hello world and a stop watch. You still will not understand the relative capacity for any particular piece of work once you do. This is because work always causes the CPU speed differences to be mitigated by other bottlenecks such as waiting for memory or I/O and the impact varies with workload, time, and machine architecture. The rules developed early on were based on some intuitive understanding that heavy computational work is less impacted by non processor bottlenecks, whereas transactional workloads with lots of locking and data sharing are more impacted. Since most other machines come from a heritage which emphasizes the former and zSeries heritage is firmly in the latter, the intuitive choices are generally correct and don't change with normal evolutionary changes in processor speed. Joe Temple [EMAIL PROTECTED] 845-435-6301 295/6301 cell 914-706-5211 home 845-338-8794 Alan Altmark/Endicott/To: [EMAIL PROTECTED] [EMAIL PROTECTED]cc: Sent by: Linux onSubject: Re: Perpetuating Myths about the zSeries 390 Port [EMAIL PROTECTED] IST.EDU 10/29/2003 02:00 PM Please respond to Linux on 390 Port On Wednesday, 10/29/2003 at 10:08 PST, Jim Sibley [EMAIL PROTECTED] wrote: I disagree about your I/O wait is independent of processor speed because that is only one component of reponse time. The faster processor can initiate I/O's faster and can service interrupts faster, thus reducing internal queue wait times. The faster processor can start more I/Os per second than a slower processor. A faster I/O processor can move data off the channel into memory faster (and vice versa). The speed of the channel itself and the device does not change. Yes, you CAN change it, but it is a function (and price) that is independent of CPU selection. As far as transaction transit time, that's a
Re: ECKD-less system?
The first card in the 1130 deck actually did the entire cold start. You could operate the machine after but had to know the control panel. Of course its always easier with some software. I had a friend who could cold start an 1130 faster than you could take the card off the top of the reader put it in the hopper and press start. He did this entirely by manipulating the switches on the front panel. Talk about arcane skills... Joe Temple [EMAIL PROTECTED] 845-435-6301 295/6301 cell 914-706-5211 home 845-338-8794 Tom Duerbusch [EMAIL PROTECTED]To: [EMAIL PROTECTED] net cc: Sent by: Linux onSubject: Re: ECKD-less system? 390 Port [EMAIL PROTECTED] IST.EDU 09/24/2003 12:38 PM Please respond to Linux on 390 Port In LPAR, couldn't you still IPL from a tape drive? Now, if there was only an Escon attached reader. Just think of the card deck and resulting floor sort I remember IPL'ing an IBM 1130 via cards. Just one hand full. The Suse Linux 8.1 starter system is over 20,000 cards (2 cases, 5 boxes per case). If you actually had the entire system on cards, could you ever, actually, IPL? Tom Duerbusch THD Consulting Arnd Bergmann wrote: On Wednesday 24 September 2003 16:50, Adam Thornton wrote: I haven't tried this, so I might be totally smoking crack, but: Is there any reason you couldn't build the FCP driver into the kernel, and then IPL from an NSS. You'd need an ECKD-based system to build the inital NSS but then once it was done I don't see why you couldn't IPL from NSS (no initrd, of course) and mount all your disks as SCSI. Correct. You could also IPL from VM virtual reader devices or from a shared dasd device when using the VM load parameter patch. Note that the your IPL disk needs to have the kernel on it but not necessarily the root file system. Of course, none of this works on LPAR installations and VM always needs DASD devices to run on itself. Arnd
Re: InfoWorld Article - Microsoft Benchmarks Step Up Linux Assault
Also, if you move a bunch of intel servers onto the faster intel server it still probably still is being run at low or very low utilization and how many customers would want to put that many eggs in basket that is has an Intel server's reliability? Joe Temple [EMAIL PROTECTED] 845-435-6301 295/6301 cell 914-706-5211 home 845-338-8794 Jim Sibley [EMAIL PROTECTED]To: [EMAIL PROTECTED] hoo.com cc: Sent by: Linux onSubject: Re: InfoWorld Article - Microsoft Benchmarks Step Up Linux Assault 390 Port [EMAIL PROTECTED] IST.EDU 09/05/2003 01:59 PM Please respond to Linux on 390 Port I wonder who in IBM made this statement! IBM has no plans to begin running Linux performance benchmarks for the mainframe, the IBM spokesman said. We typically don't run industry standard performance benchmarks for any software on the mainframe, he said. The value of the mainframe is based on its ability to securely run multiple applications on a single platform as opposed to purely seeking outstanding performance of one application on a single platform. He obviuosly does not understand the power of the mainframe. IBM has always talked about throughput rather than performance - elapsed time of any given transaction and throughput. The real questions are: Could the zSeries give a the same or higher transaction rate? Hang a terabyte on the a zSeries 116 or 216, give it 64 GB memory, and see how much work you can put through it. (The zSeries 116 or 216 is in the range of the same clock rate). How many Microsoft/Intel's would it take to do the same work! And what would the cost/transaction be? = Jim Sibley Implementor of Linux on zSeries in the beautiful Silicon Valley Computer are useless.They can only give answers. Pablo Picasso __ Do you Yahoo!? Yahoo! SiteBuilder - Free, easy-to-use web site design software http://sitebuilder.yahoo.com
Re: Whither consolidation and what then?
Tzfir Cohen wrote And if you had all of those Office machines as separate images on a giant T-Rex, those IT folks would still have to manually patch each and every image separately, and spend 15 minutes on that. As for cloning, patch distribution etc.: those solutions are exactly solutions (?) to the management problem. As you mentioned in the beginning, just cramming many images on one mainframe won't make it go away. One of the interesting things about total cost is that centralization/consolidation surfaces costs that are otherwise hidden. I certainly don't need to have admin functions for my laptop nickel and diming my time, but it never shows up on the books. The inexpensiveness of PC's is one of the myths of IT that go right along with the inflated cpu speed myths about the mainframe... VM does provide an opportunity to do somewhat better with cloning solutions by allowing the files which make an image to be shared. These files need only be updated once for a set of images. My view is that there is a trade off between cost to consolidate images and the cost to manage images. This says that for each application/situation there is an optimum amount of image consolidation which is driven ability of zLinux to scale up, the ability of the application to scale up, and the skill and mindset of the application integration programmers, v the ability of the system programmers/sysadmins to exploit VM and Linux cloning tools for automation. Basically, you are looking for the minimum number of images that can be run at the availability you want, and won't cost you the farm to get to. I would also expect the number of images per unit of work to shrink over time as this optimum level is sought. This should be true whether you are working with blades or VM/linux or some combination of both. Joe Temple [EMAIL PROTECTED] 845-435-6301 295/6301 cell 914-706-5211 home 845-338-8794
Re: Performance question
While reliability is not the same as availability, availability depends on it. The less reliable the component pieces are the more redundancy is required to be available.It depends on what your target availability is. Joe Temple [EMAIL PROTECTED] 845-435-6301 295/6301 cell 914-706-5211 home 845-338-8794 Alan Cox [EMAIL PROTECTED]To: [EMAIL PROTECTED] u.org.ukcc: Sent by: Linux onSubject: Re: Performance question 390 Port [EMAIL PROTECTED] IST.EDU 06/23/2003 03:25 PM Please respond to Linux on 390 Port On Llu, 2003-06-23 at 20:22, Peter Flass wrote: Actually, if availability is critical, run on a mainframe. Everything on a mainframe is duplicated (or more), and nearly everything is hot-swappable. If one power supply fails, the other takes over. If a DASD fails, RAID recovers without a hiccup. You probably need two OSA adapters attached to two phone lines, preferably to two central offices, but that's about all. Mainframe is a very expensive way to get availability, and a very poor one at the extreme end. Its a very good way of getting reliability and the two are quite different.
Re: Processor Comparisons
It takes somewhere between 1 and 4 900 MHz SPARC III to do the work of a z900T (around 900 MHz) engine, DEPENDING on the workload. This is at equal utilization on both machines. Use 1.5 to 2 if you don't know what the workload is. We use the following rules for utilization: Benchmark 100% Head to Head Production (Single Boxes) 60% Other machine duplexed for availability 40% Hot Backup for other machine 30% Production Server Consolidation 25% Server Consolidation with Test, QA, etc. 15% ISV 10% So a server consolidation case would be 6-8 SPARC's per Turbo. Lots of caveates, lots of complicating factors but this will get you a start. Get your IBM rep for a sizing. They should be able to get local FTSS assistance for these kinds of estimates and help to get the utilization measured, etc. Joe Temple [EMAIL PROTECTED] 845-435-6301 295/6301 cell 914-706-5211 home 845-338-8794 Gerard Graham [EMAIL PROTECTED]To: [EMAIL PROTECTED] cc: Sent by: Linux onSubject: Processor Comparisons 390 Port [EMAIL PROTECTED] IST.EDU 03/27/2003 10:23 AM Please respond to Linux on 390 Port Has anyone gotten anywhere with processor comparisons ? I am now being asked to give estimates such as what is the IBM equivalent in MIPs to a Sun E4500/Solaris 8 CPU box or a Sun 4200 Solaris, I myself look at these machines and have no idea how big or small they are. I am not talking utilization or what apps are running on it I am just trying to get some kind of grid for comparisons. I am constantly being asked these questions and people are tired of the canned it depends answer. It is difficult to move forward with the Linux on z/series without some kind of ballpark numbers. Again ballpark I can't give a potential customer the how a CPU works answer they will just look away and go and by another server. Any help is appreciated. - This message and its attachments may contain privileged and confidential information. If you are not the intended recipient(s), you are prohibited from printing, forwarding, saving or copying this email. If you have received this e-mail in error, please immediately notify the sender and delete this e-mail and its attachments from your computer.
Re: Interesting perspective
John Summerfield writes If the application stays up, ir's more reliable... ... I'm sure that's actually true in IBM mainframes too. I read recently a new disk drive from IBM, I guess in many respects a successor to RAID. A disk failed? Leave it there, swap in a spare. Zero maintenance because failed components are swapped out of service, spares swapped in. Are the individual drives especially reliable? No. Is the storage device especially reliable? Yes. Does anyone care about the fine distinction? No. If the application stays up its more available. If it were more reliable you would not have to take any action (ie spend money) to repair the failed part. This is where the distinction is, and to the extent that the repairs cost time and money, people care. This is what drives the idea of autonomic computing; the more the systems can repair themselves, the more availability turns into reliability. Clusters are a long way from autonomic today. I guess my main points were that clusters in and of themselves do not make reliability irrelevant, many clusters have less availability than we might think at first blush, and they all lose availability as the load on them grows. And yes this is true for all systems, including zSeries. But zSeries does have advantages that come from Virtualization, the balanced machine structure, the autonomic features built in and the high reliability of the hardware. Joe Temple [EMAIL PROTECTED] 845-435-6301 295/6301 cell 914-706-5211 home 845-338-8794 John Summerfield [EMAIL PROTECTED]To: [EMAIL PROTECTED] afe.com.au cc: Sent by: Linux on 390Subject: Re: Interesting perspective Port [EMAIL PROTECTED] EDU 03/19/2003 06:34 AM Please respond to Linux on 390 Port On Tue, 18 Mar 2003, Joseph Temple wrote: I would point out that clustering makes hardware more available, not more reliable. If the application stays up, ir's more reliable. The things actually fail more often because there is more to fail, I'm sure that's actually true in IBM mainframes too. I read recently a new disk drive from IBM, I guess in many respects a successor to RAID. A disk failed? Leave it there, swap in a spare. Zero maintenance because failed components are swapped out of service, spares swapped in. Are the individual drives especially reliable? No. Is the storage device especially reliable? Yes. Does anyone care about the fine distinction? No. but the user sees the cluster as available during the failures. There are some problems with availability clustering as it is usually done, causing the cluster to have lower (often significantly lower availability) than it is designed to have. The first problem is that as the utilization rises on a cluster the redundancy in the cluster drops, unfortunately so does the reliability of the components. Systems whose load is growing start losing availability from day one as the workload grows. Most folks add hardware when they need it for capacity, not when they need it for availability. Next when running with one or more redundant servers down, the probability of failure of the remaining servers increases due to stress brougth on by higher utilization. Because of this n+1 availability is not usually a good enough design point. The second problem is that utilization must be kept quite low if to maintain redundancy unless the utilization grows linearly with load. If the through put tails off or saturates as the utilization goes up, the utilization required to maintain redundancy is lower than intuitively expected. Most people don't have a clue about how their workload saturates on a cluster, let alone at what utilization they loose the redundancy required to get the availability they desire. Furthermore, n+2 availability is met a lower utilizations than n+1 availability. The third problem is that failover time is often long enough to count as a measurable outage, particularly when a data base or shared state is involved. As far as I know the IBM Parallel Sysplex with data sharing and redundant coupling facilities is the only system that can avoid a measurable outage on a failover. In today's multiple tiered systems the availability of the tiers is mulitplied, so that the availability of the whole solution is somewhat less than the availability of the weakest tier. Finally, the Linux on z Solution has a an advantage on patching in that multiple virtual machines can share the unpatched and patched versions. You only have to update the shared image once and then roll the boot/ipl of the VMs to point to the new version. In addition the virtual machines' redundant capacity can be handled by letting the remaining machines have
Re: Interesting perspective
Tzafrir Cohen said:Yes, but if you bring clustering into the game, then suddenly cheaper hardwares can become more relieble. The author also forgets that the guests need patching as well. Having all of them as guests on a mainframe, or as separate machines in a farm is not all that different in that respect, because remote-management tools are good enough for the basic tasks. And you can still load the new software to one unused computer in the farm, start it, and then swap-out the bad computer you want to retire. Requires some more hardware, but the hardware is much cheaper, anyway. A bigger problem is that there are simply more machines to patch. This is the basic issue: machines are not patched because their admins (or admin-replacements) don't bother. Admining a system is not a task that requires a special admin (that should be aware of patching). I would point out that clustering makes hardware more available, not more reliable. The things actually fail more often because there is more to fail, but the user sees the cluster as available during the failures. There are some problems with availability clustering as it is usually done, causing the cluster to have lower (often significantly lower availability) than it is designed to have. The first problem is that as the utilization rises on a cluster the redundancy in the cluster drops, unfortunately so does the reliability of the components. Systems whose load is growing start losing availability from day one as the workload grows. Most folks add hardware when they need it for capacity, not when they need it for availability. Next when running with one or more redundant servers down, the probability of failure of the remaining servers increases due to stress brougth on by higher utilization. Because of this n+1 availability is not usually a good enough design point. The second problem is that utilization must be kept quite low if to maintain redundancy unless the utilization grows linearly with load. If the through put tails off or saturates as the utilization goes up, the utilization required to maintain redundancy is lower than intuitively expected. Most people don't have a clue about how their workload saturates on a cluster, let alone at what utilization they loose the redundancy required to get the availability they desire. Furthermore, n+2 availability is met a lower utilizations than n+1 availability. The third problem is that failover time is often long enough to count as a measurable outage, particularly when a data base or shared state is involved. As far as I know the IBM Parallel Sysplex with data sharing and redundant coupling facilities is the only system that can avoid a measurable outage on a failover. In today's multiple tiered systems the availability of the tiers is mulitplied, so that the availability of the whole solution is somewhat less than the availability of the weakest tier. Finally, the Linux on z Solution has a an advantage on patching in that multiple virtual machines can share the unpatched and patched versions. You only have to update the shared image once and then roll the boot/ipl of the VMs to point to the new version. In addition the virtual machines' redundant capacity can be handled by letting the remaining machines have the resources of the VM that was rolled out which it then are reclaimed on restart. The hardware utilization stays relatively constant because workloads saturate zSeries machines less than other machines. This is because saturation comes from non processor bottlenecks and the zSeries machines are more robust in supply of other resources per CPU configured. As a result a virtual cluster will see higher redundancy at any utilization and therefore will be more available than the equivalent cluster, EVEN IF THE CLUSTER HARDWARE were AS RELIABLE AS zSERIES, which it is not. Joe Temple [EMAIL PROTECTED] 845-435-6301 295/6301 cell 914-706-5211 home 845-338-8794
Re: URGENT! really low performance. A related question...
Lucius wrote Forgive me for being dumb here, but I'd like to ask How? If you're sharing a VM minidisk among several Linux guests, how can you update the contents without having all of the guests brought down? You start with 2 LPARs and 2 VMs each has a set of shared Linux disks at whatever level. Each can use the full capacity of the machine (Sharing engines, channels, OSAs, etc.). When you take one side down to upgrade it the other takes on the load. You can use one VM and 2 sets of Linux disks and share memory if constrained inthat manner, but then VM outages take you down. The outage time is essentially a failover time, which depends on what you do with the data. Being an IBM z guy, I would say that to maximize uptime the data belongs in a zOS data sharing sysplex, where the state and lock structures are kept in redundant coupling facilities, essentially eliminating the failover time there, but any failover mechanism can be used. Joe Temple [EMAIL PROTECTED] 845-435-6301 295/6301 cell 914-706-5211 home 845-338-8794 Lucius, Leland Leland.Lucius@ecTo: [EMAIL PROTECTED] olab.comcc: Sent by: Linux onSubject: Re: URGENT! really low performance. A related question... 390 Port [EMAIL PROTECTED] IST.EDU 02/19/2003 11:39 AM Please respond to Linux on 390 Port With disk sharing and VM, the apparent outage for maintenance of Linux can be virtually eliminated. Forgive me for being dumb here, but I'd like to ask How? If you're sharing a VM minidisk among several Linux guests, how can you update the contents without having all of the guests brought down? Thanks, Leland
Re: URGENT! really low performance. A related question...
Alan Cox wrote Was it alpha or gamma emitters they got in their materials ? It was alpha particles and it was not unique to IBM. Basically it started with dynamic RAM. A passing particle drains charge from the memory cell causing soft errorsECC became mandatory for dynamic ram when the 64K bit chips were introduced because the charge held was small enough that the drainage changed the cell's state. The IBM 8130 was the first machine to ship with 64Kbit chips (yes that's K) and we had to scramble to retrofit ECC into the design. The 4381 shipped around the same time and had similar problems. As things got smaller static memory also started to be affected and we started to see ECC on caches as well as mainstorage. I don't know this for a fact but I suspect Sun's L2 cache problems were related to soft errors. One other wrinkle, IBM's use of flip chip did make the problem more pronounced because the active chip area was on the side closed to the substrate which is an emitter. The wire bond technique used by other vendors mitigated but did not eliminate the problem, because the emitted particiles had to get through the whole chip to hit the memory cells. However the soft error rate still indicated the use of ECC, particularly as memory got denser. Joe Temple [EMAIL PROTECTED] 845-435-6301 295/6301 cell 914-706-5211 home 845-338-8794 Alan Cox [EMAIL PROTECTED]To: [EMAIL PROTECTED] u.org.ukcc: Sent by: Linux onSubject: Re: URGENT! really low performance. A related question... 390 Port [EMAIL PROTECTED] IST.EDU 02/20/2003 10:23 AM Please respond to Linux on 390 Port On Thu, 2003-02-20 at 01:00, John Alvord wrote: And Lord protect you if the packaging accidently contained materials which generated gamma rays. Another tale of woe from the IBM 1980s Gamma seems odd, it doesn't interact much most times, now alpha emitters I could believe. Was it alpha or gamma emitters they got in their materials ?
Re: URGENT! really low performance. A related question...
John Summerfield wrote Why do people keep referring to the speed of light? In what I learned of electronics, signals are carried by electrons travelling round in conductors (and semiconductors). AFAIK electrons are quite a bit slower than photons. Well, according to a guy named James Clerk Maxwell light is electromagnetic radiation. Electrical current (the flow of charge) in a conductor is induced by an electromagntic wave. At low frequency and short distance the idea of voltage and current works fine and the wave is ignored. As things get faster or longer (ie POwer Lines) the conductors need to be treated as wave guides more than as conductors. As far as the math goes, I believe that you have to start using transmission line characteristics as soon as the delay on the line matters. The conductors in a modern chip are not treated as simple wires with no impedance or delay but are modeled with inductance, capacitance, and resistance, in much the same way that a transmission line is modeled. An upper bound for electromagnetic wave speed is C, unless you get into some really hairy quantum physics paradoxes. (Read Shroedingers Kittens, I forget the author's name) On the other hand practical physical limitations slows waves down. How close to C you get depends on the medium and practical things like the need to dampen reflections on the line. That is the fastest line is useless if the signal on it rings enough to prevent further use of the line. Wow I thought I had forgotten all that stuff 30 years ago when I burned my fields and waves book... Joe Temple [EMAIL PROTECTED] 845-435-6301 295/6301 cell 914-706-5211 home 845-338-8794
Re: URGENT! really low performance. A related question...
John Summerfield wrote That tells me you weren't current on your maintenance... Software currency is an issue, but there is very good reason why people bring their systems down only once or twice a year for maintenance. They lose money when they do it. The balance between having the right fixes on and keeping the system up is an art. Of course it is also possible in a sysplex to put maintenance on without taking a system wide outage.I know of a bank that has a weekly maintenance cycle but has kept their sysplex up for more than 5 years. While sysplex is a Z/OS thing similar things can be done with LPAR, VM and some relatively simple failover scripts. There remain a few hardware and microcode updates which require that a box be taken down, but such maintenance is relatively rare and usually is not urgent. Security alerts for VM and Z/OS are practically nonexistent, and it is not necessary to take down VM to do maintenance on the Linux systems. With disk sharing and VM, the apparent outage for maintenance of Linux can be virtually eliminated. This is one of the key elements of TCO. Joe Temple [EMAIL PROTECTED] 845-435-6301 295/6301 cell 914-706-5211 home 845-338-8794 John Summerfield summer@computerdatasTo: [EMAIL PROTECTED] afe.com.au cc: Sent by: Linux on 390Subject: Re: URGENT! really low performance. A related question... Port [EMAIL PROTECTED] EDU 02/18/2003 07:59 PM Please respond to Linux on 390 Port I just IPL'ed the S/390 Sunday 2/9/03 it was up since we installed our new MP3000 1/9/02 that's January 9, 2002. I IPLed to install Z/VM 4.3.0 (Scheduled Change) That tells me you weren't current with your maintenance;-) If you looked at the security advisories and decided they were not needed, that's fine. However, I suspect that many people who report how long *their* systems have been up have neglected their maintenance. -- Cheers John Summerfield Microsoft's most solid OS: http://www.geocities.com/rcwoolley/ Note: mail delivered to me is deemed to be intended for me, for my disposition. == If you don't like being told you're wrong, be right!
Re: URGENT! really low performance. A related question...
John Summerfield wrote I presume, from what you say, that Java isn't all that wonderful on zSeries? Improved CPU performance may make it so. One cannot make such blanket statements. JAVA is a language, not a workload. Yes it does have characteristics that cause it to have long path lengths. However, it also has characteristics that trash caches, particularly if the programmer takes OO programming seriously. Small caches get trashed faster than large caches particularly when they are in fast engines. The balance of pathlength and cache misses is entirely dependent upon the application in any language. Java just happens to be less efficient on all fronts than earlier languages, but then Fortran is less efficient than assembler. I would argue that the slide in code efficiency is balanced by the increase in processor speed over time for all machines. Relative capacity is more related to how the programmer writes the application and how much compute v data is involved. Long ago we used to call the ratio of Execution to Bandwidth the E/B ratio. This ratio still applies, when E/B is large the other machines will look better than z. When it is small the z shines. This is true regardless of language. Joe Temple [EMAIL PROTECTED] 845-435-6301 295/6301 cell 914-706-5211 home 845-338-8794
Re: URGENT! really low performance. A related question...
Mark Drvodelsky wrote But the question still does not appear to be answered - why does the mainframe have to run at such a low clock speed? The answer to your question has to do with how chip real estate is used. In a zSerires micro processor the primary usage of area is for large L1 caches and error detection/recovery hardware. Basically, increases in cache size result in decreases in clock rate. This is because there is more load on the critical signals. Secondly, to date the zSeries microprocessor pipleline does not do super scalar processing. That is it finishes 1 instruction per cyle at best. This is because it takes consideratbly more work and hardware to do mainframe style error recovery functions when more than 1 instruction can complete in a cyc;le. While super scalar execution does not help with clock speed it does help with cpu intense measurements like SPECint. However, since the cache is larger the zSeries will wait for memory less often than other machines.Metrics like SPECint and MHz ignore cache misses. So the question becomes how much are the caches missing? The more they miss the better the zSeries looks. This is very workload dependent. One driver of cache misses is context switches; another is I/O. If you attempt to make an Intel server very busy, the cache miss rate will climb, causing throughput to saturate, unless the work is very CPU intense and cache working set per transaction or per user is very small. The reason the Robert Nix's print server dabacle occured is that IBM made the mistake of treating Samba file/print as a single type of workload. We didn't understand at the time that a print server can behave like a network to network prototcol server. These servers actually move very little data through the cpu. Such a machine has very little context switching and the I/O is network to network which will actually drive very little data through the caches. The combination makes the workload cpu intense and if busy a bad candidate for Linux/z. By contrast a Samba file server can be doing enough disk to network I/O which pushes more data through the caches changing blocks to packets. This can cause distributed servers can get I/O and cache bound. Samba can be either CPU or I/O intense, and the single context makes the cpu intense workloads unattractive for z particularly if the machines are busy. So the answer to your question is that we could build a zSeries microprocessor which is as fast as any other processor, but to do so would cause us to lose the fundamental strengths in context switching, data caching and I/O. There is alwasy a trade off between speed and capacity. zSeries favors capacity; Intel favors speed. How much L1 cache should be given up to increase the clock rate? How much RAS and recovery function should be given up to improve SPECint? We have seen this situation improve over time, and IBM will continue to improve its microprocessor design, but zSeries cannot simply abandon strength in large working set workloads to crank up the clock speed and/or instruction rate for workoads with small working sets. This particularly true when the virtualization and workload management which drive consolidation and mixed workloads is dependent on the very hardware capabilities that would have to be given up. Joe Temple [EMAIL PROTECTED] 845-435-6301 295/6301 cell 914-706-5211 home 845-338-8794 Mark Darvodelsky Mark_Darvodelsky@royalTo: [EMAIL PROTECTED] sun.com.aucc: Sent by: Linux on 390 Subject: Re: URGENT! really low performance. A related question... Port [EMAIL PROTECTED] U 02/16/2003 08:32 PM Please respond to Linux on 390 Port But the question still does not appear to be answered - why does the mainframe have to run at such a low clock speed? Perhaps someone with some hardware knowledge could explain it? Why can't the clock be cranked up to be the same speed as the latest Pentium? Most of us mainframe guys understand its inherent advantages, but as someone has already commented, it often just doesn't wash with management if a cheap Pentium outperforms a million-dollar mainframe. Regards. Mark Darvodelsky Data Centre - Mainframe Facilities Royal SunAlliance Australia Phone: +61-2-99789081 Email: [EMAIL PROTECTED] CAUTION - This message is intended for the addressee named above It may contain privileged or confidential information. If you are not the intended recipient of this message you must not use, copy, distribute or disclose it to anyone other than the addressee. If you have received this message in error please return the message to the sender by replying to it and then delete the message from your computer. Internet emails are not
Re: URGENT! really low performance.
Robert Nix wrote: But, if one image starts doing compiles or compression of large quantities of data, or any other CPU bound task, everyone will suffer. Actually you have a choice. If the compiles, etc. are relegated to a compute server you can make it suffer rather than everyone else, also, if you cap the cpu given the guests you can minimize the intensity of t the suffering when cpu heavy tasks occur, but it will go on for a longer period of time. It's a matter of prioities and how you distribute work among virtual machines. The beauty of Linux is that the compute intense server can be a virtual or real machine, but it is still LInux.In the past such a scheme using reeal machines would split the work between ZOS and WIndows which is a lot more complex. We need to start thinking about things like Grids of virtual and real servers. Joe Temple [EMAIL PROTECTED] 845-435-6301 Nix, Robert P. Nix.Robert@mayo.To: [EMAIL PROTECTED] edu cc: Sent by: Linux onSubject: Re: URGENT! really low performance. 390 Port [EMAIL PROTECTED] IST.EDU 02/13/2003 04:01 PM Please respond to Linux on 390 Port Mainframes do I/O exceptionally well, but when it comes to compute bound tasks, they do very poorly. If you think about a tar operation, the compression is a fairly compute-intensive operation. We're running a 9672-R56 w/ one IFL. During our initial trial, we found the IFL to be about the same as a 300 or 400mHz PC for compute-bound tasks. The strength of the mainframe comes in for burst-type execution and I/O throughput. Things like multiple web servers running in individual Linux images. File serving. Anything where: A) The CPU isn't expected to be taxed a great deal. and B) the CPU isn't going to be utilized for long periods of time. This allows the CPU to be shared among a larger quantity of images, giving all of them the impression of a dedicated box. But, if one image starts doing compiles or compression of large quantities of data, or any other CPU bound task, everyone will suffer. Robert P. Nixinternet: [EMAIL PROTECTED] Mayo Clinic phone: 507-284-0844 RO-CE-8-857page: 507-270-1182 200 First St. SW Rochester, MN 55905 Codito, Ergo Sum In theory, theory and practice are the same, but in practice, theory and practice are different. -Original Message- From: Alex Leyva [SMTP:[EMAIL PROTECTED]] Sent: Thursday, February 13, 2003 3:10 PM To: [EMAIL PROTECTED] Subject: URGENT! really low performance. Hi all, i have a problem, we have a z800, the configuration is: 1 cp 80 MIPS 1 IFL 8 Gb storage 3 partitions: -os/390 2.6 -os/390 2.6 -z/vm 4.3 840 gb (shark) the cp is dedicated to both os/390, and the ifl to z/vm, 2gb to both os/390, and 6 gb to z/vm. Redhat 7.2 as a z/vm guest: [root@linux1 root]# uname -a Linux linux1.xxx.xxx.xxx 2.4.9-38lvm #1 SMP mii feb 12 12:25:01 CST 2003 s390 unknown [root@linux1 s390]# cat /proc/cpuinfo vendor_id : IBM/S390 # processors: 1 bogomips per cpu: 630.78 processor 0: version = FF, identification = 02900A, machine = 2066 [root@linux1 s390]# cat /proc/meminfo total:used:free: shared: buffers: cached: Mem: 1045737472 364187648 6815498240 15532032 317743104 Swap: 4098211840 409821184 Default installation, the z/vm has one week installed: q cplevel z/VM Version 4 Release 3.0, service level 0201 (64-bit) Generated at 05/09/02 17:30:26 EST IPL at 02/07/03 12:13:53 EST when we make a tar -gzipping it- from a directory with 100Mb, we have that: -the hmc indicates that the ifl is at 99% utilization. -real time monitor indicates that the processor is at 99% utilization: | USERID %CPU %CP %EM ISEC PAG WSS RES UR PGES SHARE VMSIZE TYP,CHR,STAT | | LINUX1 99 .15 99 4.4 .00 100K 100K .01 50%A 1G VUS,QDS,DISP | | SYSTEM.08 .08 .00 .00 .000 5060 .0 536 . 2G SYS, | | VMRTM .02 .01 .01 .63 .00 462 483 .00 3%A32M VUS,IAB,SIMW | -top at the linux shows: 30 processes: 27 sleeping, 3 running, 0 zombie, 0 stopped CPU states: 97.6% user, 2.3% system, 0.0% nice, 0.0% idle Mem: 1021228K av, 279636K used, 741592K free, 0K shrd, 14120K buff Swap: 400216K av, 0K used, 400216K free 234992K cached we apply some performance related commands like: set quickdsp linux1 on real set share linux1 relative 300 real set share linux1 absolute 50% real and the time went from 1m3.6s to 1m2.039s in the better case, the people from ibm (they are here yet) can give me an answer
Re: VM for Intel?
I apologize for not following the whole thread here, but in case it has not been mentioned, the following should be pointed out to further differentiate z from x as far as virtualization goes: The zSeries architecture and hardware design contains facilities not found in the Inel machine. These facitlies (mainly SIE and EMIF virtual memory related architectures go a long way in reducing the overhead fo virutalization which must be done entirely in software on Intel. Furthermore the small caches and relatively high aggregate memory latency of the intel machines means that they suffer more from the increase in context switching that occurs when virtualization is done. So yes they can do it, but not nearly as well. Joe Temple [EMAIL PROTECTED] 845-435-6301 295/6301 cell 914-706-5211 home 845-338-8794 -- Forwarded by Joseph Temple/Poughkeepsie/IBM on 02/20/2002 07:57 AM --- Jim Elliott [EMAIL PROTECTED]@VM.MARIST.EDU on 02/19/2002 08:50:15 PM Please respond to Linux on 390 Port [EMAIL PROTECTED] Sent by:Linux on 390 Port [EMAIL PROTECTED] To:[EMAIL PROTECTED] cc: Subject:Re: VM for Intel? But that was my question. Since IBM and VMWare are partnering on this effort, would IBM have contributed any sort of functionality lifted from z/VM? If not, why the partnership? ... Mark: Just as IBM supports Linux across all four of our server lines, we also wanted to support partitioning across all four of our server lines. With this announcement (which we previewed at LinuxWorldExpo in NYC), we now have that function available. There is no z/VM code in VMware's ESX server (what IBM will be shipping on selected xSeries servers). Regards, Jim Elliott - Linux Advocate, IBM Canada