I will shot gun some of them... 1. Disaster recover is much easier on the mainframe. In effect, no matter what hardware is replaced with what hardware, it is all the same. With PC type servers, the hardware, hence the software drivers are constantly changing. This may force you to reinstall the software instead of just restoring. 2. Our I/O subsystem. Mainframes with ficon/FCP, can drive (per IBM documentation) drive hundreds of thousands of I/Os per second. If you only need a few hundred I/Os per second, well, that is within PC ranges. 3. Licensing is a two edge sword. Putting 5 copies of Oracle on an IFL...you only pay for one copy. However, if you have many, one copy products, you end up needing more engines on the IFL, which (if you get charged by the engine), causes those product charges to increase. 4. Disk is disk. It costs the same whether your DS8100/DS6800 is configured for CKD or SCSI disk. 5. Mainframe memory is more expensive, but it is more effectively used. When an application states that it needs 4 GB to run, I start around 500 MB and increase it when needed. 6. When a server application needs more resources, many times you have to go out and buy a newer, bigger server. When a mainframe server needs more resources, you may have options to rob other servers. Also for larger shops, Capacity on Demand. 7. Green. There is an application on z10s and above, that will show you your footprint and the incremental footprint for additional loads. I seem to recall something about you can plug in data from servers you are migrating from, to show the incremental decrease in the footprint of the datacenter. There hasn't been much chatter about this on the listservs so I don't know how well this has been received. 8. Internal network speed. If a function requires the use of several servers and they are network attached, things are slowed up by the network. No such problem with Hypersockets or Guest Lans/VSWITCH (under VM) and can have large packets also. 9. We don't, but we should have performance tools. You buy one for the LPAR and you know what is going on. Rather than buy one per server. You still might need specialized performance tools on some servers. Oracle OEM to measure internal Oracle performance, for example. 10. The serious problem with PC servers is context switching. There a dog. Mainframes are great at this, as CICS transactions really drive this. If your load tends towards transactional instead of batch (data mining), PC type servers were not designed for this. I assume that RS6000 and Sun type servers are pretty good at context switching, but I have no direct knowledge of this. Back when Linux started hitting mainframes and IFLs were announced, there was discussions of 100 images per engine. A lot of the servers at that time were routers, DNS, Samba, NFS and some web. Now I seem to here 10-20 real workloads per engine. BTW, there was/is an MES upgrade from one box to another. In the case of the MES upgrade from a z/890 (our box) to a z10 (hopefully/maybe ours), the license for the IFLs transfers. Which means that we would not have to pay for the Linux side again. And the new IFLs are faster per engine than the older IFLs. That is no longer a cost on the mainframe that you still have on the other server platforms. Know that I think of it, I may be thinking of the MES upgrade that pulled cards (and you license and CPUID) from one box and installed it on the newer box. I'm now thinking that the IFL engine transfer will happen with any upgrade to a new box. I've been looking at the MES upgrade option for so long, that I have MES on the mind <G>. Tom Duerbusch THD Consulting
>>> John Cousins <[email protected]> 12/6/2010 11:07 AM >>> Here we go again! Without success, we've been trying to get the IT department here to adopt z/Linux since 2003! Our zVM licence has been recently cancelled, and I have just had a request from our Enterprise Architects for some costing for z/Linux as they need to compare server virtualisation costs with VMware! One problem of trying to get a cost per virtual server was always trying to estimate how many servers an IFL will support. We had a 13 SuSe servers defined in a z800 IFL but as they were hardly used we couldn't measure a thing! So are there any rules of thumb out there on how many production virtual servers would run on a Z10 IFL? Obviously it will depend on server utilisation, guess that will need to be estimated as well? Another question is where do the bulk of the savings come from? From my investigations over the years other success stories suggest most savings come from software licensing, e.g Oracle, Tivoli etc. but also from networking infra-structure by the use of virtual switches. Are there any other areas that provide benefits? Any ideas or constructive suggestions would be gratefully received! Best regards John John Cousins Senior IT Officer Central Support Services ICT Division Bristol City Council Romney House Romney Avenue PO Box 1380 Bristol BS7 9TB Tel : 0117 922 4705 Fax: 0117 922 3983 e-mail: [email protected] ______________________________________________________________________ 'Do it online' with our growing range of online services - http://www.bristol.gov.uk/services Sign-up for our email bulletin giving news, have-your-say and event information at: http://www.bristol.gov.uk/newsdirect View webcasts of Council meetings at http://www.bristol.gov.uk/webcast Bristol is the UK's first Cycling City. Visit www.betterbybike.info to join thousands of others getting around by bike. ---------------------------------------------------------------------- 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 more information on Linux on System z, visit http://wiki.linuxvm.org/ ---------------------------------------------------------------------- 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 more information on Linux on System z, visit http://wiki.linuxvm.org/
