KAIO Performance question

Hello, We are testing an SCO 5.0.5 server, on a dual Pentium II Xeon, 512MB RAM and 4x9GB disks, with IDS 7.30 UC2 We have a stored procedure that takes around 90 minutes to run. Enabling or disabling KAIO does not make any difference in the 90 minutes. With KAIO enabled, we make better use of the dual processors (38% idle average). With KAIO disabled, we might as well run on one processor (49% idle average). Without KAIO, we do not expect Informix to make a good use of the second processor as the stored procedure is mainly OLTP. We would expect better usage of the second processor with KAIO as unix is sharing the load. So, the question is: should we just accept that KAIO does not provide better performance on SCO, or are we missing some tuning parameter? BTW, we have tried the same hardware with NT instead of SCO (IDS 7.3 TC7), with the same database and the same disk layout. Guess what: our procedure runs in exactly the same time. Bashar Chalabi CTL, London

In article <930166299.14913.0.nnrp-08.c29fbb28@news.demon.co.uk>, "Bashar Chalabi" <bashar@ctl.com> wrote: How much CPU Vps you are using ? Each CPU Vp runs one kio thread, so, adding CPU Vps adds kio threads and improves situation. And how much disk I/O is performed by your procedure ? Is disk used hardly ? -- With best regards, Yuri Dovgart, SAP R/3, Informix consultant, "Telecominvest" company. E-mail y_dovgart@tci.ukrtel.net > Hello, > > We are testing an SCO 5.0.5 server, on a dual Pentium II Xeon, 512MB RAM and > 4x9GB disks, with IDS 7.30 UC2 > > We have a stored procedure that takes around 90 minutes to run. > > Enabling or disabling KAIO does not make any difference in the 90 minutes. > > With KAIO enabled, we make better use of the dual processors (38% idle > average). With KAIO disabled, we might as well run on one processor (49% > idle average). > > Without KAIO, we do not expect Informix to make a good use of the second > processor as the stored procedure is mainly OLTP. We would expect better > usage of the second processor with KAIO as unix is sharing the load. > > So, the question is: should we just accept that KAIO does not provide better > performance on SCO, or are we missing some tuning parameter? > > BTW, we have tried the same hardware with NT instead of SCO (IDS 7.3 TC7), > with the same database and the same disk layout. Guess what: our procedure > runs in exactly the same time. > > Bashar Chalabi > CTL, London > > Sent via Deja.com http://www.deja.com/ Share what you know. Learn what you don't.