Oninit cdi / Archive
Talk to an expert →

CPU-VP issue

Posted in 2008

Topics: Licensing & Editions, Platform-Specific Issues

STEPHEN SCOTT — — source: IIUG Forums & Mailing Lists

We have the following: IBM Informix Dynamic Server Version 10.00.FC6 HP-UX om4db1 B.11.11 U 9000/800 4126011100 unlimited-user license 3 duel core 1-gig PA-RISC procs 12 gig memory We were attempting to tune our system and changed the following: from VPCLASS cpu,num=5,noage to VPCLASS cpu,num=8,noage Once we did that the system just seemed to thrash (100% cpu utilization) for 75% of the day. Once our users went home the system worked nicely for nightly batch processing but once the users came back in the system went to junk again. Do we need to limit the number of cpu vp's to the number of physical processors? Any ideas. Stephen

↪ replying to STEPHEN SCOTT
MIKE MAGIE — — source: IIUG Forums & Mailing Lists

Straight from the doc: "On a single-processor computer, allocate only one CPU virtual processor. On a multiprocessor computer, do not allocate more CPU and user-defined virtual processors, combined, than there are CPUs on the computer." Mike

↪ replying to MIKE MAGIE
Art S. Kagel (Oninit) — — source: IIUG Forums & Mailing Lists

MIKE MAGIE wrote: > Straight from the doc: > > "On a single-processor computer, allocate only one CPU virtual processor. On a > multiprocessor computer, do not allocate more CPU and user-defined virtual > processors, combined, than there are CPUs on the computer." > Mike, the general wisdom is that this recommendation from the manuals is naive in the face of GHZ+ processor cores. The working rule of thumb seems to be that you can have 2 CPU VPs per core for any processor core faster than about 500MHZ (750 MHZ for Intel and PA-RISC). Folk running 1.5GHZ processors and faster have had good results configuring even 3 or4 VPs per core. However, you also have to factor in that multiple core processors are NOT equivalent to two or more full speed cores and adjust accordingly. The PA-RISC multi-cores are a good example as each core tests out as barely 60% of an equivalent single core processor. See my reply to the OP. Art S. Kagel > Mike > > > ******************************************************************************* > Forum Note: Use "Reply" to post a response in the discussion forum. > > See you at the IIUG Informix 2008 Conference > The Power Conference for Informix Professionals > April 27 - 30, 2008 Marriott Overland Park (Kansas City), Kansas > http://www.iiug.org/conf > Registration Now Open!! > > ================================================================================ =========== > Please access the attached hyperlink for an important electronic communications disclaimer: > > http://www.oninit.com/home/disclaimer.php > > ================================================================================ =========== > > > ================================================================================ =========== Please access the attached hyperlink for an important electronic communications disclaimer: http://www.oninit.com/home/disclaimer.php ================================================================================ ===========

↪ replying to STEPHEN SCOTT
Art S. Kagel (Oninit) — — source: IIUG Forums & Mailing Lists

STEPHEN SCOTT wrote: > We have the following: > IBM Informix Dynamic Server Version 10.00.FC6 > HP-UX om4db1 B.11.11 U 9000/800 4126011100 unlimited-user license > 3 duel core 1-gig PA-RISC procs > 12 gig memory > > We were attempting to tune our system and changed the following: > from VPCLASS cpu,num=5,noage > to VPCLASS cpu,num=8,noage > > Once we did that the system just seemed to thrash (100% cpu utilization) for > 75% of the day. Once our users went home the system worked nicely for nightly > batch processing but once the users came back in the system went to junk > again. > On 1GHZ PA-RISC processors I wouldn't configure more than 1 or 2 CPU VPs per core so that would limit you to 3-6 CPU VPS - so 8 was rather pushing the limit. However, the behavior when the number of connections goes up versus the behavior under batch load overnight would indicate that something else is going on. Post the following output and we'll see if there's anything there: Time in fractional hours since stats were last zero'd of after the last restart: ((sysmaster:sysshmval.curtime - sysmaster:sysshmval.sh_pfclrtime)/3600.00) onstat -p onstat -c onstat -g seg onstat -g glo onstat -g aio onstat -g ses Art S. Kagel Oninit > Do we need to limit the number of cpu vp's to the number of physical > processors? > > Any ideas. > > Stephen > > > ******************************************************************************* > Forum Note: Use "Reply" to post a response in the discussion forum. > > See you at the IIUG Informix 2008 Conference > The Power Conference for Informix Professionals > April 27 - 30, 2008 Marriott Overland Park (Kansas City), Kansas > http://www.iiug.org/conf > Registration Now Open!! > > ================================================================================ =========== > Please access the attached hyperlink for an important electronic communications disclaimer: > > http://www.oninit.com/home/disclaimer.php > > ================================================================================ =========== > > > ================================================================================ =========== Please access the attached hyperlink for an important electronic communications disclaimer: http://www.oninit.com/home/disclaimer.php ================================================================================ ===========

Related threads