Re: Buffer Waits Ratio

I'm guessing this formula must be: ratio = ( bufwaits/(pagreads+bufwrits) ) * 100 --parens added If so then my ratio is as low as 1%. Is this correct? onstat -p from yesterday: Informix Dynamic Server Version 7.30.UC6 -- On-Line -- Up 1 days 04:39:55 -- 4 01976 Kbytes Profile dskreads pagreads bufreads %cached dskwrits pagwrits bufwrits %cached 1419818 11786423 114203130 98.76 1530290 2316873 8113571 81.14 isamtot open start read write rewrite delete commit rollbk 190919664 1103080 24040152 25997507 581033 1697351 55587 408334 0 gp_read gp_write gp_rewrt gp_del gp_alloc gp_free gp_curs 0 0 0 0 0 0 0 ovlock ovuserthread ovbuff usercpu syscpu numckpts flushes 0 0 0 17731.75 1751.58 119 562 bufwaits lokwaits lockreqs deadlks dltouts ckpwaits compress seqscans 6 26274377 0 0 101 68524 44 ixda-RA idx-RA da-RA RA-pgsused lchwaits 363703 0 10247 352094 9244 My calculations: ratio = 185365 / (11786423 + 8113571) * 100 = ( 185365 / 19899994 ) * 100 = .009314 * 100 ratio = 0.93 % I reset my statistics daily, here is onstat -p from several days ago: Informix Dynamic Server Version 7.30.UC6 -- On-Line -- Up 2 days 18:48:08 -- 4 26552 Kbytes Profile dskreads pagreads bufreads %cached dskwrits pagwrits bufwrits %cached 12930652 30826338 492343445 97.37 5767282 6608408 29924515 80.73 isamtot open start read write rewrite delete commit rollbk 759783556 4550587 95203794 119260881 2184572 6093209 258763 1408026 4 gp_read gp_write gp_rewrt gp_del gp_alloc gp_free gp_curs 0 0 0 0 0 0 0 ovlock ovuserthread ovbuff usercpu syscpu numckpts flushes 0 0 0 80125.26 7341.69 533 1076 bufwaits lokwaits lockreqs deadlks dltouts ckpwaits compress seqscans 4002975 11 101778925 0 0 650 238254 4951 ixda-RA idx-RA da-RA RA-pgsused lchwaits 7126836 13091 63725 6586573 53663 This is more in the expected range: ratio = (4002975 / (30826338 + 29924515) ) *100 = (4002975 / 60750853) * 100 = 0.0658 * 100 = 6.58% Could some one please confirm my calculations? Thanks, Steve Murray DBA, Associated Grocers, Inc stephen_murray@agsea.com chris_burton@blue-skies.net (Chris Burton) on 11/25/99 08:47:06 AM Please respond to chris_burton@blue-skies.net (Chris Burton) To: informix-list@iiug.org cc: (bcc: Stephen Murray/AGInc) Subject: Re: Buffer Waits Ratio On Thu, 25 Nov 1999 12:33:33 +0000, Andrew Bland <blanda@hep.bradford.gov.uk> wrote: >Hi Everyone, > > I've been trying to get this ratio down below 10% I've been using a >calculation posted on here by Art. > > ratio = bufwaits/pagreads+bufwrits * 100 > > at the moment the best result I've had has been 11% when I opted for >71 LRU's .. Howver, when I was using 95 LRU's it seemed to settle on >14%. What other factors do I need to take into account? > > > I was having the same problem, and the answer I kept getting was - MORE BUFFERS. Well, it worked. But I had to increase the memory on the machine - .5 Gig to 1.5 Gig. Sometimes you don't have that luxury. On the read ahead. Check your read ahead percent used (# Used / (other three read values)). If it is >= 98%, I would rather live with buff waits at 11% because of the values of reading info from buffers that read ahead give you. If buff waits start climbing, then ease back on the read ahead. Maybe decreasing the threashold parameter by itself first, then the read ahead pages. I get good results working with one parameter at a time. I also did the recommended 127 LRU's. That seems to work pretty well too. Another important factor is to trend this info (buff waits and read aheads) over time. Maybe a week, or more. Don't bounce the engine every day trying to tune these parameters. It usually doesn't work that easily. Chris

In article <81mf91$km3$1@news.xmission.com>, Stephen_Murray@agsea.com wrote: > > > I'm guessing this formula must be: > ratio = ( bufwaits/(pagreads+bufwrits) ) * 100 > --parens added Correct with parens. > If so then my ratio is as low as 1%. Is this correct? Could be, you either have a very quiet or a very well tuned engine at least in terms of buffer/lru contention. > onstat -p from yesterday:> > Informix Dynamic Server Version 7.30.UC6 -- On-Line -- Up 1 days 04:39:55 > -- 4 > 01976 Kbytes > > Profile > dskreads pagreads bufreads %cached dskwrits pagwrits bufwrits %cached > 1419818 11786423 114203130 98.76 1530290 2316873 8113571 81.14 > > isamtot open start read write rewrite delete commit > rollbk > 190919664 1103080 24040152 25997507 581033 1697351 55587 408334 0 > > gp_read gp_write gp_rewrt gp_del gp_alloc gp_free gp_curs > 0 0 0 0 0 0 0 > > ovlock ovuserthread ovbuff usercpu syscpu numckpts flushes > 0 0 0 17731.75 1751.58 119 562 > > bufwaits lokwaits lockreqs deadlks dltouts ckpwaits compress seqscans > 6 26274377 0 0 101 68524 44 > > ixda-RA idx-RA da-RA RA-pgsused lchwaits > 363703 0 10247 352094 9244 > > My calculations: > > ratio = 185365 / (11786423 + 8113571) * 100 > = ( 185365 / 19899994 ) * 100 > = .009314 * 100 > ratio = 0.93 % The bufwaits value is not what I see above in the onstat report but maybe it just got lost in the copy. Otherwise the calculation is correct. > I reset my statistics daily, here is onstat -p from several days ago: > > Informix Dynamic Server Version 7.30.UC6 -- On-Line -- Up 2 days 18:48:08 > -- 4 > 26552 Kbytes > > Profile > dskreads pagreads bufreads %cached dskwrits pagwrits bufwrits %cached > 12930652 30826338 492343445 97.37 5767282 6608408 29924515 80.73 > > isamtot open start read write rewrite delete commit > rollbk > 759783556 4550587 95203794 119260881 2184572 6093209 258763 1408026 4 > > gp_read gp_write gp_rewrt gp_del gp_alloc gp_free gp_curs > 0 0 0 0 0 0 0 > > ovlock ovuserthread ovbuff usercpu syscpu numckpts flushes > 0 0 0 80125.26 7341.69 533 1076 > > bufwaits lokwaits lockreqs deadlks dltouts ckpwaits compress seqscans > 4002975 11 101778925 0 0 650 238254 4951 > > ixda-RA idx-RA da-RA RA-pgsused lchwaits > 7126836 13091 63725 6586573 53663 > > This is more in the expected range: > > ratio = (4002975 / (30826338 + 29924515) ) *100 > = (4002975 / 60750853) * 100 > = 0.0658 * 100 > = 6.58% > > Could some one please confirm my calculations? That also looks right to me. [SNIP] Art S. Kagel Sent via Deja.com http://www.deja.com/ Before you buy.