RE: Performance dive over time

It sounds like you may be encountering Informix defect #115327. Index pages can be incorrectly shrunk setting both leaf and node flags which will cause excess pages to be placed in the MED_HIGH buffer pool priority. (The fix deals with how the buffer manager sets the priority on the pages with 0xd0 type flag.) When performance gets bad, run the following command: onstat -P | tail -8 If the resulting output shows a large pct of the pages as "Btree", this is likely your problem. For example: $ onstat -P | tail -8 Totals: 131000 113313 17087 600 0 432 Percentages: Data 13.04 Btree 86.50 Other 0.46 One workaround is "export NOLRUPRIO=1" before starting your engine. It disables an IDS 7.3x feature, which assigns a "med-high" priority to some index pages (as shown by "onstat -R"). Newer IDS releases support a different environment variable "LRUAGE". Before using the NOLRUPRIO environment variable, review your release notes and ensure that the following defect has been fixed: 111681 - NOLRUPRIO does not work and can hang the system. I believe that it was fixed in IDS 7.31.UC4. (I can attest that it was not fixed in IDS 7.31.UC3 on AIX). -----Original Message----- From: ktroxel@my-deja.com [mailto:ktroxel@my-deja.com] Sent: Thursday, November 09, 2000 08:56 To: informix-list@iiug.org Subject: Performance dive over time Hi - I've got an instance of IDS (7.31.UC4) running on Sun E4000/Solaris 2.6. Average 320 users, 18 Gig of space. Mostly OLTP w/some statistical reports run on occasion. Over time, the performance of Informix takes a dive. It seems to degrade a little every day until it becomes unacceptable. I bounce the engine (about once a month now) and everything gets better, until over time I have to do it again. Am I missing something obvious here? Any clue would be great help. TIA Sent via Deja.com http://www.deja.com/ Before you buy.