Re: Could someone give me some direction?

On Fri, 25 Jun 1999 16:55:45 -0400, "Carlson@WHSmith" <carlson1@bellsouth.net> wrote: >See comments below . . . > >Chris Burton wrote: >> >> I have an IBM F50 with 512Mb, 12Gb database on Raid 5. It is version >> 7.30.UC7-1 on AIX 4.2.1. I am getting a lot of bufwaits. I have been >> moving the LRU's up (like Art suggests). I am currently at 20% >> bufwaits, and it has been that way regardless of the number of LRU's. >> I started with 8 and have tried 25, 35, 55 ,65, and 75 with little or >> no difference. I know that the stats show it has only been up 2 days, >> but even as time goes on (week or two) it doesn't get any better. >> Virtual memory on the system looks good - mostly used with no extra >> segments. There are about 100 sessions in the system during the day. >> It is an OLTP system (7am - 5pm - with some nightly cron jobs), and >> checkpoints are <10 seconds - mostly 5-7 seconds. Of course, all >> suggestions are welcome. >> > >I'm doing some testing now for general system performance. It was >induced by a large number of bufwaits after running many report jobs >(4gl). I've found that if BUFFERS isn't set high enough that extra LRUs >may not help much. I dropped my (test) bufwaits from 100% to 20% just >by increasing BUFFERS. See below for additional stuff. > >...snipped... >> BUFFERS 30000 # Maximum number of shared buffers> >BUFFERS could be bigger . . . about 50000 I just decreased from 33000 to 30000, because (vmstat from original post) I only had 1000-1200 free pages of memory at peak times. Now I have about 4000 pages. I was getting more paging (vmstat -s), but that has slowed down now that I reduced the size of the engine. > >> NUMAIOVPS 1 # Number of IO vps>Running KAIO??? Yes - KAIO. > >> LRU_MAX_DIRTY 10 # LRU percent dirty begin cleaning>> limit >> LRU_MIN_DIRTY 5 # LRU percent dirty end cleaning limit> >Maybe reset to 4 / 2. I will try this. > >> RA_PAGES 8 # Number of pages to attempt to read>> ahead >> RA_THRESHOLD 4 # Number of pages left before next>> group > >Depending on the speed of the disks, maybe increase these numbers. But >I'm not sure about RAID5 . . . Increasing these put bufwaits higher. > > >Just a few things to think about . .. > >John Carlson >Informix DBA >WHSmith USA Chris

See below . . . Chris Burton wrote: > > On Fri, 25 Jun 1999 16:55:45 -0400, "Carlson@WHSmith" > <carlson1@bellsouth.net> wrote: > > >See comments below . . . > > > >Chris Burton wrote: > >> > >> I have an IBM F50 with 512Mb, 12Gb database on Raid 5. It is version > >> 7.30.UC7-1 on AIX 4.2.1. I am getting a lot of bufwaits. I have been > >> moving the LRU's up (like Art suggests). I am currently at 20% > >> bufwaits, and it has been that way regardless of the number of LRU's. > >> I started with 8 and have tried 25, 35, 55 ,65, and 75 with little or > >> no difference. I know that the stats show it has only been up 2 days, > >> but even as time goes on (week or two) it doesn't get any better. > >> Virtual memory on the system looks good - mostly used with no extra > >> segments. There are about 100 sessions in the system during the day. > >> It is an OLTP system (7am - 5pm - with some nightly cron jobs), and > >> checkpoints are <10 seconds - mostly 5-7 seconds. Of course, all > >> suggestions are welcome. > >> > > > >I'm doing some testing now for general system performance. It was > >induced by a large number of bufwaits after running many report jobs > >(4gl). I've found that if BUFFERS isn't set high enough that extra LRUs > >may not help much. I dropped my (test) bufwaits from 100% to 20% just > >by increasing BUFFERS. See below for additional stuff. > > > >...snipped... > >> BUFFERS 30000 # Maximum number of shared buffers> > > >BUFFERS could be bigger . . . about 50000 > > I just decreased from 33000 to 30000, because (vmstat from original > post) I only had 1000-1200 free pages of memory at peak times. Now I > have about 4000 pages. I was getting more paging (vmstat -s), but > that has slowed down now that I reduced the size of the engine. > Paging isn't a good thing . . . I didn't know that there were any other applications on the system. > >> RA_PAGES 8 # Number of pages to attempt to read> >> ahead > >> RA_THRESHOLD 4 # Number of pages left before next> >> group > > > >Depending on the speed of the disks, maybe increase these numbers. But > >I'm not sure about RAID5 . . . > > Increasing these put bufwaits higher. I guess it depends on the amount of BUFFERS that you have available. I seemed to have an increase in bufwaits when I turned RA_* off. John Carlson Informix DBA WHSmith USA