RE: Performance Monitoring/Tuning

You need to run UPDATE STATISTICS in the manner specified in the release notes. There are several tools including Informix's ISA 1.20, which can be used to generate most of the commands for you. The LRU_MIN_DIRTY and LRU_MAX_DIRTY parameters directly impact checkpoint durations. Large numbers can cause most of your writes to be deferred until checkpoint time. Run "onstat -F". * LRU Writes are performed based on LRU_MIN/MAX_DIRTY * Chunk Writes are performed during checkpoints. -----Original Message----- From: Thomas Stainer To: informix-list@iiug.org Sent: 6/15/00 11:06 AM Subject: Re: Performance Monitoring/Tuning i rerun UPDATE STATISTICS every night - what about UPDATE STATISTICS MEDIUM or HIGH? initially my checkpoints took 0 - 2 seconds, meanwhile checkpoint duration is between 10 and 30 secs. as far as the query plan is concerned i'm not shure it is really good. "Carlson@WHSmith" wrote: > Thomas Stainer wrote: > > > > As the users of our advertisement system are moaning about the bad > > performance, i'm trying to find out why the system's perfomance isn't as > > good as it was in the beginning. Even after dropping and reimporting the > > whole database the performance was only slightly better. > > . . . and I'm presuming that you reran the required UPDATE STATISTICS? > What ONCONFIG changes were made initially?? > > > I'm running > > informix 7.30 UC6 on hp-ux 10.20, the rootdbs is located on the system's > > root-hd, the aps-dbspaces are lying in an AutoRaid (Raid5) striped over > > 8 discs - therefor fragmentation of the big tables won't better the > > performance (or am i wrong)?. > > Depends, I guess. Is performance bad because the disks are busy or > because the query plan is bad? > > > Does a change of LRU_MAX_DIRTY/LRU_MIN_DIRTY (actually 50/60) improve > > performance? > > When it comes to writing dirty pages to disk. How long are your > checkpoints?? > > > The server isn't only db-server but also application server, how many of > > the (4) CPUs should by affinated? > > Didn't know that you could do that on HPUX . . . . > > > I somewhere read, the number of LRUs should equal the number of CPUs of > > the db-server - is this right? > > I believe that may be the 'official' Informix documentation version, but > it really depends on the number of BUFFERS that you have. If buffer > contention exists, then you might need to increase the number of LRUs. > > > How can I generally monitor system's performance/throughput? > > I use glance and various onstat commands. > > > Furthermore the workload of the lun the dbspace are located on often > > averages some 80 - 90 p.c. - may be this a reason for the bad > > performance? > > > > Depends on what the performance issues are. I've seen disks at 80-90% > activity with decent response time. > > > thanx for help > > > > If you would, could you please be a bit more specific concerning what > the 'performance issues' are? > > -- > John Carlson > Informix DBA > WHSmith USA > > #include std_disclaimer.h /* These are my opinions, not my company's > opinion */ -- ------------------------------------------------- Thomas Stainer - tstainer@styria.com open-it informationsberatungsges.m.b.h. & co kg Schonaugasse 64 TEL: +43-316-875-3045 8010 Graz, Austria FAX: +43-316-875-3034