Re: linux swap to dcss?
Bill Bitner said: ... many things ... I'd completely ignore this post. It's obviously a forgery and can't possibly be from who it's supposed to be. After all, it doesn't have the words it depends in it. Rod (cough,cough, it's Friday, cough, cough) -- 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
linux swap to dcss?
PLEASE PROVIDE A SUBJECT LINE of relevance I almost ignored this and have no idea the original thread. Isn't the problem with dcss is that they are very limited? They must be less than 2GB in address - pretty hard to have a 5GB pages space fit below 2gb virtual. Lots of virtual machines already pushing way above 2gb now, i need a consistent approach. So for oracle, sap, weblogic, informix, i can't use dcss for paging. And are the rules for paging out vdisk different than for dcss? The issue with vdisk is that cp considers a dedicated vdisk as shared so is slow to page it out. That could be fixed but paging out dcss seems slower now than for vdisk. So i think real storage will be more constrained using dcss for paging than using vdisk. except of course for the ibm benchmark machines with over configured storage so that would not be detected in lab measurements but would be of great concern in real life??? Date: Thu, 18 Nov 2004 14:13:27 +0100 From: Carsten Otte [EMAIL PROTECTED] On Nov 17, 2004, at 10:32 AM, Carsten Otte wrote: You might as well be interrested in our DCSS execute-in-place filesystem. This one allows you to put your applications and shared libraries in DCSS segments, and they can be used/executed on individual guests without copying into virtual guest storage. Hasn't Rob van der Heij also shown that swap in EW segments is even faster than swap-on-vdisk? I haven't tried this yet but it ought to be fairly straightforward to set up at boot time. Yea, you're telling me news I told Rob before [circle closed]. The point is: I do not recommend it to customers until we are done testing. But looks to be very much faster than vdisk indeed. with kind regards Carsten Otte If you can't measure it, I'm Just NOT interested!(tm) // Barton Robinson - CBW Internet: [EMAIL PROTECTED] Velocity Software, IncMailing Address: 196-D Castro Street P.O. Box 390640 Mountain View, CA 94041 Mountain View, CA 94039-0640 VM Performance Hotline: 650-964-8867 Fax: 650-964-9012 Web Page: WWW.VELOCITY-SOFTWARE.COM // -- 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: linux swap to dcss?
Hi Barton, today you have a full 2G of space for dcsses available - even when using lage main memory. You can use dcss swapping with oracle, and all other apps you have in mind. You can mix different swapping devices easily with Linux, you can even priorize using them. When you need 500gig of swap, you can use 2gig dcss high prio, and 498gig on dasds. The VM paging priority issue you describe does not apply to DCSS based swap. What's your point? with kind regards Carsten Otte -- omnis enim res, quae dando non deficit, dum habetur et non datur, nondum habetur, quomodo habenda est -- 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: linux swap to dcss?
On Nov 18, 2004, at 10:04 AM, Carsten Otte wrote: Hi Barton, today you have a full 2G of space for dcsses available - even when using lage main memory. I may be confused, but I thought you had to define your DCSSes above the top of the memory Linux used. What does the IPL line look like if you want 100 MB at 1G, and space usable by Linux both above and below it? Adam -- 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: linux swap to dcss?
In the case of swapping to EW DCSS it would be a separate address space so the rules are not the same as the XIP2 type of mapped file system. David From: Linux on 390 Port on behalf of Adam Thornton Sent: Thu 11/18/2004 11:15 AM To: [EMAIL PROTECTED] Subject: Re: linux swap to dcss? On Nov 18, 2004, at 10:04 AM, Carsten Otte wrote: Hi Barton, today you have a full 2G of space for dcsses available - even when using lage main memory. I may be confused, but I thought you had to define your DCSSes above the top of the memory Linux used. What does the IPL line look like if you want 100 MB at 1G, and space usable by Linux both above and below it? Adam -- 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: linux swap to dcss?
On Thu, 18 Nov 2004 11:20:48 -0500, David Kreuter [EMAIL PROTECTED] wrote: In the case of swapping to EW DCSS it would be a separate address space so the rules are not the same as the XIP2 type of mapped file system. No you're wrong. The DCSS must be mapped into the virtual machine address space somewhere under the 2G. The trick that Carsten refers to is have Linux support a 'gap' in memory under the 2G where the segments can be mapped - whether for swap or xip2. -- Rob van der Heij rvdheij @ gmail.com -- 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: linux swap to dcss?
Hi Carsten, i just don't believe that under stress conditions that in today's environement you can show value to paging to dcss over vdisk. This being with z/vm 5.1 and a current redhat or suse. The ibm recomendations i keep hearing being repeated are that swap sizes must be some multiple of the linux virtual machine size. So if my Oracle server requires 2GB, then (some) ibm do says i should have 4GB of swap. and in today's environement, i can not have a discontigous virtual machine, so swap dcss must be at the top of virtual storage. but that would be over the 2gb line. not possible - with what is available or announced as far as i know. mixing dcss and vdisk seems to create a more complex environment than most installations want to utilize. after all, vm is just a hypervisor i'm told. even if discontigous machines were supported by linux and z/vm, i still don't have 2gb available of virtual storage for the dcss, unelss you eliminate all need for linux to address storage below the 2gb line? In real production environements, i currently have a serious problem with storage and minimizing storage requirements. i believe that with the current vdisk and dcss page stealing algorithms, that dcss will retain more real storage than vdisk. much more. and i think vdisk could be greatly improved upon. so really, my main point i'm trying to make is that a real measurement showing value is worth a lot of opinions i have mine, you have yours, rob has his, let's duke it out on the benchmark battlefield show real data winner buys the beer? From: Carsten Otte [EMAIL PROTECTED] Hi Barton, today you have a full 2G of space for dcsses available - even when using lage main memory. You can use dcss swapping with oracle, and all other apps you have in mind. You can mix different swapping devices easily with Linux, you can even priorize using them. When you need 500gig of swap, you can use 2gig dcss high prio, and 498gig on dasds. The VM paging priority issue you describe does not apply to DCSS based swap. What's your point? with kind regards Carsten Otte If you can't measure it, I'm Just NOT interested!(tm) // Barton Robinson - CBW Internet: [EMAIL PROTECTED] Velocity Software, IncMailing Address: 196-D Castro Street P.O. Box 390640 Mountain View, CA 94041 Mountain View, CA 94039-0640 VM Performance Hotline: 650-964-8867 Fax: 650-964-9012 Web Page: WWW.VELOCITY-SOFTWARE.COM // -- 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: linux swap to dcss?
I may be confused, but I thought you had to define your DCSSes above the top of the memory Linux used. What does the IPL line look like if you want 100 MB at 1G, and space usable by Linux both above and below it? This issue has been solved. With our new memory detection you can use #cp def store config to create the memory as you like to. Linux does autodetect this with a new enhanced memory detection. DCSSes can be loaded into a hole created before. example: 0..128meg storage, 2gig...10gig storage = you can load dcsses between 128meg and 2gig. So far this is not in a customer distribution, but you can use a Vanilla 2.6. kernel from kernel.org to play with it. with kind regards Carsten Otte -- omnis enim res, quae dando non deficit, dum habetur et non datur, nondum habetur, quomodo habenda est -- 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: linux swap to dcss?
In the case of swapping to EW DCSS it would be a separate address space so the rules are not the same as the XIP2 type of mapped file system. David Nope, you can do the same with xip2fs. You can create storage holes and fill them with your filesystem. with kind regards Carsten Otte -- omnis enim res, quae dando non deficit, dum habetur et non datur, nondum habetur, quomodo habenda est -- 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: linux swap to dcss?
On Thu, 18 Nov 2004 17:52:13 +0100, Carsten Otte [EMAIL PROTECTED] wrote: Nope, you can do the same with xip2fs. You can create storage holes and fill them with your filesystem. For the oldbies among us - this is effectively what we did with CMS and segment reserve, because we did not have def stor config in those days. -- Rob van der Heij rvdheij @ gmail.com -- 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: linux swap to dcss?
Hi Carsten, i just don't believe that under stress conditions that in today's environement you can show value to paging to dcss over vdisk. This being with z/vm 5.1 and a current redhat or suse. Why don't you measure it? My simple measurements (I am not a performance guy but the developer) indicate a strong benefit. The ibm recomendations i keep hearing being repeated are that swap sizes must be some multiple of the linux virtual machine size. So if my Oracle server requires 2GB, then (some) ibm do says i should have 4GB of swap. and in today's environement, i can not have a discontigous virtual machine, so swap dcss must be at the top of virtual storage. but that would be over the 2gb line. not possible - with what is available or announced as far as i know. mixing dcss and vdisk seems to create a more complex environment than most installations want to utilize. after all, vm is just a hypervisor i'm told. Sles9 will allow to have discontiguous storage very soon. So does Rhel4. And Vanilla Linux does have it since 2.6.6. Your statement is clearly outdated. even if discontigous machines were supported by linux and z/vm, i still don't have 2gb available of virtual storage for the dcss, unelss you eliminate all need for linux to address storage below the 2gb line? That's true. You need to have approx 64megs stoage below, but that's not a significant amount. In real production environements, i currently have a serious problem with storage and minimizing storage requirements. i believe that with the current vdisk and dcss page stealing algorithms, that dcss will retain more real storage than vdisk. much more. and i think vdisk could be greatly improved upon. Discussion with the VM development team indicates that things are just the other way round. They believe DCSS would be more elegant for their page stealing algos - unless I got them completely wrong. so really, my main point i'm trying to make is that a real measurement showing value is worth a lot of opinions i have mine, you have yours, rob has his, let's duke it out on the benchmark battlefield show real data winner buys the beer? I would really love to prove my point by a performance measurement, but I believe people are much better off when I keep developing good soloutions that integrate Linux and VM instead of trying to prove my points by performance measurements. Maybe you should visit one of either Rob's or my presentations to learn about current development. with kind regards Carsten Otte -- omnis enim res, quae dando non deficit, dum habetur et non datur, nondum habetur, quomodo habenda est -- 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: linux swap to dcss?
So ibm would rather an outside vendor do the measurements? and that any customer with performance problems should run future or unsupported levels of code in production? And that vm developement's opinions are better than real data? my my. Date: Thu, 18 Nov 2004 18:12:50 +0100 From: Carsten Otte [EMAIL PROTECTED] Hi Carsten, i just don't believe that under stress conditions that in today's environement you can show value to paging to dcss over vdisk. This being with z/vm 5.1 and a current redhat or suse. Why don't you measure it? My simple measurements (I am not a performance guy but the developer) indicate a strong benefit. The ibm recomendations i keep hearing being repeated are that swap sizes must be some multiple of the linux virtual machine size. So if my Oracle server requires 2GB, then (some) ibm do says i should have 4GB of swap. and in today's environement, i can not have a discontigous virtual machine, so swap dcss must be at the top of virtual storage. but that would be over the 2gb line. not possible - with what is available or announced as far as i know. mixing dcss and vdisk seems to create a more complex environment than most installations want to utilize. after all, vm is just a hypervisor i'm told. Sles9 will allow to have discontiguous storage very soon. So does Rhel4. And Vanilla Linux does have it since 2.6.6. Your statement is clearly outdated. In real production environements, i currently have a serious problem with storage and minimizing storage requirements. i believe that with the current vdisk and dcss page stealing algorithms, that dcss will retain more real storage than vdisk. much more. and i think vdisk could be greatly improved upon. Discussion with the VM development team indicates that things are just the other way round. They believe DCSS would be more elegant for their page stealing algos - unless I got them completely wrong. If you can't measure it, I'm Just NOT interested!(tm) // Barton Robinson - CBW Internet: [EMAIL PROTECTED] Velocity Software, IncMailing Address: 196-D Castro Street P.O. Box 390640 Mountain View, CA 94041 Mountain View, CA 94039-0640 VM Performance Hotline: 650-964-8867 Fax: 650-964-9012 Web Page: WWW.VELOCITY-SOFTWARE.COM // -- 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: linux swap to dcss?
So, one developer stating that his time is better spent writing code than doing performance testing for which he's not qualified, somehow represents an official IBM position? I think you're stretching a little here, and shooting the messenger on top of it. Mark Post -Original Message- From: Linux on 390 Port [mailto:[EMAIL PROTECTED] On Behalf Of Barton Robinson Sent: Thursday, November 18, 2004 12:38 PM To: [EMAIL PROTECTED] Subject: Re: linux swap to dcss? So ibm would rather an outside vendor do the measurements? and that any customer with performance problems should run future or unsupported levels of code in production? And that vm developement's opinions are better than real data? my my. -- 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: linux swap to dcss?
So ibm would rather an outside vendor do the measurements? and that any customer with performance problems should run future or unsupported levels of code in production? And that vm developement's opinions are better than real data? my my. You really seem to try hard to get me wrong. We are off cause doing performance measurements, but because we do things professional we wait for our distribution partners to include our new stuff in their distributions. Our performance people do publish measurements on internal builds, that would not reflect the customers view! As I stated in an earlier posting, we do not recommend DCSS swapping to be used in production until we're done testing it. Once we do recommend it, quite frankly it is supported as well. And when it's tested and supported, yes: run it in production then! I did not state that I do not want to look at performance data in the end, but I do not want to publish results from my simple initial measurements on development code. I want to wait for our performance group to finish their professional measurements. As I indicated in earlier posting, both our theory about how things work as well as our measurements we have so far indicate a very strong advantage in using DCSS for swapping over VDISK. Go pick up the code from kernel.org and measure it, or just stop complaining. By-the-way: Where is your proof for your statement? with kind regards Carsten Otte -- omnis enim res, quae dando non deficit, dum habetur et non datur, nondum habetur, quomodo habenda est -- 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: linux swap to dcss?
A few thoughts on various statements: 1.The Linux development process is a little different than the traditional S390 where we test things ad nauseum before we let them out the door. I apologize that VM Performance hasn't had time to fully measure the swap to dcss. It's on the list, but there are a few more pressing issues right now. But this discussion is good; it helps me think about what to consider in the measurements. 2.A DCSS segment defined as EW will have a lower priority in the VM storage management steal algorithms, unlike traditional segments and vdisks. So it doesn't have the same challenge as vdisk currently does. 3.Swapping to DCSS can be done as a block device which can lower the pathlength (i.e. not having to create CCWs etc.) compared to vdisk swapping. 4.Using recent VM and Linux you can create gaps for the DCSS, but the DCSS is still limited below 2GB. So it won't be an acceptable alternative in all cases. 5.The swapping itself can occur without VM intervention, except for the creation and when required the actual paging of the DCSS (that's VM paging, not Linux). Bill Bitner - VM Performance Evaluation - IBM Endicott - 607-429-3286 -- 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
