Re: Online 5 to 722 Performance

David Williams wrote: > In article <355807F3.54B5@mcmail.com>, Paul Watson <watson@mcmail.com> > writes > >Art S. Kagel wrote: > >> > >> Paul Watson wrote: > >> > > >> > This weekend upgrade a server from Online 5.x to 7.22.UC5. Everything went > >> > except the 722 runs 75-100% slower than 5. All the index's have been > >> > rebuilt and updates statistics have been run as per the release notes. > >> > > >> > I think I might be seeing the recent 7.23 and 7.24 Tech alert behaviours. > >> > > >> > Anybody any ideas. > >> > > >> > BTW the box is a Sun 1000/4 Processors/312MB complete with SSA. Raid is not > >> > being used, neither is mirroring. There are 3 CPUS bound to the engine, > >> > the disk accesses are all under 8ms. The thread profiles are poor though. > >> > >> For best 5.xx like performance make sure that OPTCOMPIND is set to zero > >> (0) in your ONCONFIG file! You should also go for 7.24UC5 or later or > >> 7.3UC2 or later which fixes many problems introduced in 7.22 and 7.23. > >> Also what version of Solaris are you running? Only 7.24UC5 and later > >> are certified for the latest version of Solaris. > >> > > > >Setting OPTCOMPIND from 2 to 0 makes no difference. Weird eh! After > >each engine change the 4ge was recompiled, shouldn't make a difference I > >know but ..... > > > >What I find really odd is the following > > > >Import the data, run time = 30 minutes > >Import the data, run 'update statistics', run time = 30 minutes > >Import the data, run 'update statistics high/low etc', run time = 30 > >minutes > > > >Solaris 2.5.1 and as far as I can tell all the patches are in place. > >The machine was recently setup by Sun themselves as the Sun reseller was > >unable to get the box stable. > > > >BTW the disk access times are significantly better than under 5. > >Everything seems to point to either statistics problem or memory. The > >thread profiles are about 30-40% of what I would expect to see. From > >the sqexplain there are only index reads and most of them are > >key-only. > > > > > >> Art S. Kagel > > > > How many CPUs? 4 Physical, 3 CPUVPS all bound > How much memory? 312MB 50000 Buffers initial allocation 125MB - doesn't grow > > > sar -p 1 1 > run twice shows? No paging at all > > > sar -d 1 1 > On the first run of the code I get busy Q < 5% access <8msOn the second run the disks are not accessed > onstat -g seg I don't have access to the box at the moment, but the virtual had > 1000 free pages, everything looked 'normal' onstat -g ath Again I don't the box in front of me but 3cpus 4 kaio, 4 page cleaners and again all the normal threads, oh and 1 sqlexec when the code is running > and > > onstat -g glo> Nothing much all the activity is on the cpus ie > 98% of total usage > onstat -c I'll have to get this. But SHMBASE as per release notes NUMAIOCPU = 1 LRUmax/min 15/10 SHVIRTSIZE=14000 SHMADD=8000 BUFFERS=50000 LOCKS=2000 PDQPRIORITY=0 LRUS=4 CLEANERS=4 NOAGE=0 AFF_NPROCS=3 NETTYPE=ipcshm,1,75,CPU tlitcp,1,10,NET -- Paul Watson # WF Software Ltd # If it ain't broke, Tel: +44 1436 674729 # fix it till it is. Fax: +44 1436 678693 # www.wfsoftware.co.uk #