Disappointing test results from 9.40

HP-UX 11.11 I'm doing some benchmark testing on a critical process to our client - alter fragment init ... in dbspace of medium and large tables. In every case, 9.40 (I'm trying both FC1 and FC2) is substantially (that is 20-40%) *slower* than the equivanlent statement on 9.21 FC4. I have kaio enabled in all the tests, and am using the same $ONCONFIG file for each, so the test ought to be pretty representative of the IDS versions' performance under HP-UX 11.11. Actually, I disabled kaio for one test and it made no difference. Has anyone else had a similar, or contrary, experience with IDS v9.40? thanks Neil

I have refined the testing further, and broken the alter fragment into its component parts: 1. an alter fragment with no indexes 2. a re-build of the indexes In both cases 9.40 is markedly slower than 9.21, on the same hardware, chunks and ONCONFIG parms. "Neil Truby" <neil.truby@ardenta.com> wrote in message news:bqg8jn$21e6lv$1@ID-162943.news.uni-berlin.de... > HP-UX 11.11 > > I'm doing some benchmark testing on a critical process to our client - alter > fragment init ... in dbspace of medium and large tables. In every case, > 9.40 (I'm trying both FC1 and FC2) is substantially (that is 20-40%) > *slower* than the equivanlent statement on 9.21 FC4. > > I have kaio enabled in all the tests, and am using the same $ONCONFIG file > for each, so the test ought to be pretty representative of the IDS versions' > performance under HP-UX 11.11. Actually, I disabled kaio for one test and > it made no difference. > > Has anyone else had a similar, or contrary, experience with IDS v9.40? > > thanks > Neil > >

On Tue, 2 Dec 2003 13:08:40 -0000, "Neil Truby" <neil.truby@ardenta.com> wrote: >I have refined the testing further, and broken the alter fragment into its >component parts: > >1. an alter fragment with no indexes >2. a re-build of the indexes > >In both cases 9.40 is markedly slower than 9.21, on the same hardware, >chunks and ONCONFIG parms. > > One of the things I've noticed from 9.21 is that detached index builds are much slower than attached index builds (or semi-detached index, to be technical). I've used PDQ for index builds (as part of a reorg) and that seems to speed things up a bit... Can't speak as to 1), though...

Neil, I see the same. We are using 9.40 FC2 for Solaris 2.8. A bug has been opened for my case: Index builds taking 50-300 % longer then in 7.31 regardless of amount of PDQ resources allocated. Slowness shows up with larger tables. In our case 150-800 mill rows. Support says that the engine is erroneously writing empty pages to the physical log when building the index. A fix is slated. onstat -D will confirm this behavior for 9.40 FC. You will not see this in 7.31. I also see slowness (~50 %) with onpload for large tables with same symptom. Rich "Neil Truby" <neil.truby@ardenta.com> wrote in message news:bqg8jn$21e6lv$1@ID-162943.news.uni-berlin.de... > HP-UX 11.11 > > I'm doing some benchmark testing on a critical process to our client - alter > fragment init ... in dbspace of medium and large tables. In every case, > 9.40 (I'm trying both FC1 and FC2) is substantially (that is 20-40%) > *slower* than the equivanlent statement on 9.21 FC4. > > I have kaio enabled in all the tests, and am using the same $ONCONFIG file > for each, so the test ought to be pretty representative of the IDS versions' > performance under HP-UX 11.11. Actually, I disabled kaio for one test and > it made no difference. > > Has anyone else had a similar, or contrary, experience with IDS v9.40? > > thanks > Neil > >