Bufwaits and #LRUs

Hi, The more LRUs the less bufwaits ? therefore, always 127 ? someone says 1000 buffer per LRU. I have 10000 buffers, so 10 LRU, but I have bufwaits. Is there another reason ? better to set 127 whichever #buffers ? Regards, Thanks, Manel Falcó

Manel Falc' i Aige wrote in message <399c2491.37785568@news.bcn.ttd.net>... >Hi, > >The more LRUs the less bufwaits ? therefore, always 127 ? > >someone says 1000 buffer per LRU. I have 10000 buffers, so 10 LRU, > but I have bufwaits. Is there another reason ? better to set 127 >whichever #buffers ? > I have found 127 works best even with a single processor machine. Test it and monitor your checkpoint times. >Regards, Thanks, >Manel Falc' >

> I have found 127 works best even with a single processor > machine. Test it and monitor your checkpoint times. Also 127 cleaners ? really needed ? overhead ? I have one processor, IDS 7.30UC7 Actually I have short checkpoints. From 0 to 3 seconds. MAXLRU 2, MINLRU 1. I think Jonathan doesn't recommend 1, 0 why ? Ok, Thanks, Manel Falcó On Thu, 17 Aug 2000 23:22:47 +0100, "smooth1" <smooth1@iclway.co.uk> wrote: > >Manel Falcó i Aige wrote in message <399c2491.37785568@news.bcn.ttd.net>... >>Hi, >> >>The more LRUs the less bufwaits ? therefore, always 127 ? >> >>someone says 1000 buffer per LRU. I have 10000 buffers, so 10 LRU, >> but I have bufwaits. Is there another reason ? better to set 127 >>whichever #buffers ? >> > >>Regards, Thanks, >>Manel Falcó >> >

Manel Falc' i Aige wrote in message <399ce62a.722602@news.bcn.ttd.net>... >> I have found 127 works best even with a single processor >> machine. Test it and monitor your checkpoint times. > >Also 127 cleaners ? really needed ? overhead ? I tried 127 CLEANERS when do you on onstat -u they appear next the start. Monitor at the end of 1 days run and see how many are idle (all 0's in most columns) reduce as needed, add afew for peak loads. >I have one processor, IDS 7.30UC7 >Actually I have short checkpoints. From 0 to 3 seconds. >MAXLRU 2, MINLRU 1. I think Jonathan doesn't recommend 1, 0 why ? > Probably because this would reduce the write cache percentage since dirty pages are then only cached for a few seconds. >Ok, Thanks, >Manel Falc' > >On Thu, 17 Aug 2000 23:22:47 +0100, "smooth1" <smooth1@iclway.co.uk> >wrote: > >> >>Manel Falc' i Aige wrote in message <399c2491.37785568@news.bcn.ttd.net>... >>>Hi, >>> >>>The more LRUs the less bufwaits ? therefore, always 127 ? >>> >>>someone says 1000 buffer per LRU. I have 10000 buffers, so 10 LRU, >>> but I have bufwaits. Is there another reason ? better to set 127 >>>whichever #buffers ? >>> >> >>>Regards, Thanks, >>>Manel Falc' >>> >> >

smooth1 wrote: > > Manel Falcó i Aige wrote in message <399ce62a.722602@news.bcn.ttd.net>... > >> I have found 127 works best even with a single processor > >> machine. Test it and monitor your checkpoint times. > > > >Also 127 cleaners ? really needed ? overhead ? > > I tried 127 CLEANERS when do you on onstat -u they appear > next the start. Monitor at the end of 1 days run and see how many > are idle (all 0's in most columns) reduce as needed, add afew > for peak loads. > > >I have one processor, IDS 7.30UC7 > >Actually I have short checkpoints. From 0 to 3 seconds. > >MAXLRU 2, MINLRU 1. I think Jonathan doesn't recommend 1, 0 why ? 1/0 works fine. I've been using it on various server versions for years. > Probably because this would reduce the write cache percentage > since dirty pages are then only cached for a few seconds. Writing dirty pages does not remove them from the cache it just moves them to the clean buffer pool. The write cache % will however be affected somewhat, but compared to 2/1 not much. Art S. Kagel > >Ok, Thanks, > >Manel Falcó > > > >On Thu, 17 Aug 2000 23:22:47 +0100, "smooth1" <smooth1@iclway.co.uk> > >wrote: > > > >> > >>Manel Falcó i Aige wrote in message > <399c2491.37785568@news.bcn.ttd.net>... > >>>Hi, > >>> > >>>The more LRUs the less bufwaits ? therefore, always 127 ? > >>> > >>>someone says 1000 buffer per LRU. I have 10000 buffers, so 10 LRU, > >>> but I have bufwaits. Is there another reason ? better to set 127 > >>>whichever #buffers ? > >>> > >> > >>>Regards, Thanks, > >>>Manel Falcó > >>> > >> > >