Performance-Question

Hi all, IDS 9.40 UC3 (Linux 2.4.20) I'm wondering about a weired phenomenon concerning We have a SLEECT-Statement over 4 large tables. For me it seems, that the Query-Optimizer does a good job for this query. But now we have the follwing effect: Starting this statement immediatly after a server-reboot, is takes 35 Seconds. Waiting 25 seconds, and repeating the query, it takes 1.5 seconds. Waiting some minutes (there might be some traffic on the database) it takes 10 seconds. Now my question: Is it possible, that it takes so much time to load indixes into the cache? My Buffers are set to max-values: 600.000 at 2K PageSize. The informix-docs says, that since Version 9.40 it is not possible to keep indixes or tables memory-resident because the server does it by itself. Any ideas? Thanks in adnvance! Bye Markus

Marcus, before running your slect statement, can you run an onstat -z tehn run an onstat -p after the query has completed. Can you do this for all 3 scenarios you have decribed. The post the top couple of lines of the onstat -p output , the bit that talks about dskreads , read cache5 etc Thanks "Markus Bschorer" <mb@worxbox.com> wrote in message news:<c8v0eq$r1q$02$1@news.t-online.com>... > Hi all, > > IDS 9.40 UC3 (Linux 2.4.20) > > I'm wondering about a weired phenomenon concerning > > We have a SLEECT-Statement over 4 large tables. For me it seems, that the > Query-Optimizer does a good job for this query. But now we have the follwing > effect: > > Starting this statement immediatly after a server-reboot, is takes 35 > Seconds. > Waiting 25 seconds, and repeating the query, it takes 1.5 seconds. > Waiting some minutes (there might be some traffic on the database) it takes > 10 seconds. > > Now my question: Is it possible, that it takes so much time to load indixes > into the cache? My Buffers are set to max-values: 600.000 at 2K PageSize. > The informix-docs says, that since Version 9.40 it is not possible to keep > indixes or tables memory-resident because the server does it by itself. > > Any ideas? > > Thanks in adnvance! > > Bye > Markus