Re: ONLINE 7.2 tuned to simulate SE x.x

In article <5b26mc$8b8@cssun.mathcs.emory.edu>, billy@west.co.za ("Billy Wheeler") wrote: >Hoo-boy! This was a very ill-advised decision. You don't just install >OnLine 7.2 - you should have done training or had consultants in to >advise you of the various things you can and should do to make OL >fly. > As a matter of fact, Billy, we didn't 'just install' ONLINE. Some thought did go into configuring it. We changed our UNIX parameters as specified in the release notes. We set up dbspaces on 4 disks (2 data, 1 logicals, 1 physical). 3 disk controllers are utilized for them. Mirroring is off, Log files are about 50 x 1Mb, Shared memory is resident, Buffers are at 32 Mb (1/4 of System 128MB), Single_cpu_vp = 1, locks 15000, numaiovps = 4, phsbuff = logsbuff = 32 kb, shmvirtsize = 32768, shmadd = 8192, 8 lrus, lru_max_dirty to 3, lru_min_dirty to 2 (Lru writes are still rare), stacksize to 64, ra_pages to 10 with threshold to 4, dbspacetemp is specified, max_pdqpriority is at 50 (can't let OLTP suffer), optcompind = 0. Statistics have been updated MEDIUM on databases, with HIGH on every single index COLUMN. All tables in all databases have been modified to LOCK MODE (ROW). There's a nightly onstat -F. There is no relay module. Terminal (character) connections are thru tcpip. What else are we supposed to do the get this beast to run? Hand-hold the AIOVPS thru their writes? Sing mantras to the LRU queues? Grease the BUFFERS? You tell me! >You don't mention your hardware config. but OL7 takes a *lot* more >RAM than SE. More RAM? I thought that this line was patented by MICROSOFT & INTEL? >On 8 Jan 97 at 3:08, Rudy Fernandes wrote: > >> We have recently 'upgraded' from SE 5.0 to ONLINE 7.2. Performance >> dropped dramatically on the first day (2 days back), to about 10% of >> SE performance. > >This is not unheard of. Your ONCONFIG might be totally inappropriate. > I admit, BUFFERS was inappropriate. (set to 200, then 2000, now 16000). >> A frenetic 36 hours later performance is up to about 30 to 50% of SE >> level - the users are no longer tearing their hair out, they just >> have long faces. >> >> With SE, I hardly had any configuration to do. Update Statistics >> once in a while, keep meaningful indexes, recreate them once in a >> while, make sure the programs are using them, keep the log on >> different disk and I'm in business. > >This is quite correct. > >> When using ONLINE, its a different ballgame with the myriad of >> tunable parameters. >> >> It would seem logical that a large majority of parameters of ONLINE >> are also used by SE, except that the DBA cannot tune them. > >This is *completely* false. OnLine has a completely different >internal architecture than SE. Communication between client and >server is completely different. In fact, if you didn't recompile and >are using the Delay Module, er, Relay Module, that alone could be >your problem. > >> What are they? And what are their values under SE? I can think of the following parameters which any database server could use. I would guess that in SE they would be generally be implemented at the OS level. BUFFERS (clean, at least), RESIDENT, LOCKS, STACKSIZE, RA_PAGES, RA_THRESHOLD, DBSPACETEMP, OPTCOMPIND. They seem to be the crucial ones for OL7 too. > >You should really go on the DSA System Admin course, and possibly >Performance Tuning For OnLine7. > In reply to that, @#$%*&#!$% :-) ---------------------- Rudy Fernandes GIC, Kuwait ----------------------