Jim, might I suggest you apply to the next redbook??? As for your LPAR, that probably explains a lot. In the real world, I can't figure out why a customer who has to pay for it would dedicate an LPAR to Linux that is at best today 1.3Ghz (z990), at 1-2 orders of magnitude more in price than a say 3Ghz Intel chip (With memory that is a lot less $$) and oh by the way, I recently lost a proof of concept because until we get complete FCP, the I/O on the "ix" platform was faster.
With VM, we can share the resource and utilize resources much more effectively and at higher utilization, and the price points are very different. So, there are proper platforms for each workload, and though we would like lots of it on z, some just runs better elsewhere. What z does better is not single big things, but lots and lots of little things. So we're probably in total agreement, some successful PUBLISHED analysis of current capacity would be really nice, but comparing LPAR to any other platform realistically would likely make the LPAR look expensive. (In which case you couldn't publish it, right?) One POC i've been looking at it is replacing several "boxes" with virtual machines with very detailed analysis. It would I think be better to get this published than other "benchmarketed" numbers. And I absolutely LOVE this quote from Dale: >(An other, unrelated truth, I always liked "A little experience >can ruin lot of good theory." A quote often attributed it >Enstine.) >From: Jim Sibley <[EMAIL PROTECTED]> > >Barton wrote: > >"There were two redbooks this >year that looked at many performance issues. If >anything, they were >productive in finding performance issues that needed >to be >addressed." > >I'm not addressing "tuning", but rather taking issue >with the fact that there is very little recent >performance and capacity information. A lot of >measurements were done early, s/390z Linux was >categorized, then a series of assumptions have become >the mythology of Linux on zSeries. Since then, zSeries >has changed a lot. > >One of things that happens when machines become faster >and bigger is that they enable more applications on >that platform. Until the G6, CMOS techonolgy was even >slower than the bipolar technology that it replace. >Now, the zSeries has caught up with and now exceeds >its previous capabilities spurred on by the >competition of intel and Unix processors. We need to >evaluate some of the assumptions we made early on (3 >years ago!). > >For example: August, 2003, share >http://www.linuxvm.org/present/index.html > >Thoss's report on performance is about the lastest >there is: z900 z216, F20 shark, ESCON/FICON, gigabit >ethernet, 2.4.7 or 2.4.17 kernel, 31 bit. > >Currently available to the customer: >z990, 800 shark, FICON, 2.4.21 kernel, hipersockets, >64 bit. (some of the test in Thoss's report were >memory constrained in a 31 bit environment). > >Also, >- the redbooks emphasize overall VM tuninig with >really very little information about tuning Linux - >especially if you have to run in an LPAR >- a dearth of information on things like DB2, >websphere >- lots of reports on how zseries scaled up to an >"equivalent" intel, but not really very much of what >happens when zseries scales BEYOND intel (both in CPs >and I/O) - large transactions volumes and rates. > "If you can't measure it, I'm Just NOT interested!"(tm) /************************************************************/ Barton Robinson - CBW Internet: [EMAIL PROTECTED] Velocity Software, Inc Mailing 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 /************************************************************/
