Re: AIO and Buffer Waits

Hi Alberto, if calculating the LRU queues as you suggested the value would be 2400 LRUS not 240. Are you sure that would be correct? The values AUTO_CKPTS 1 RTO_SERVER_RESTART 600 are set. I cannot increase the physical log because I have 2 GB chunks as biggest. And you cannot use more than one chunk for the physical log. The values of LRU_MAX_DIRTY and LRU_MIN_DIRTY are so little because we have 4,8 GB BUFFERS. During the checkpoints these buffers have to be written to disk. I'm afraid that the checkpoints would last too long if I increase those values and that this would produce blockages. Thanks, Reinhard. From: Alberto Romeu Pessonio Filho [mailto:arfilho@orizonbrasil.com.br] Sent: Wednesday, July 13, 2011 3:01 PM To: Habichtsberg, Reinhard Subject: RES: AIO and Buffer Waits Reinhard: Using the concepts from the old school, i suggest you to calculate between 750 and 1000 buffers fer LRU queue. Also, try to enable AUTO_CKPTS=1 and RTO_SERVER_RESTART =1800 on your onconfig to able your instance to work with RTO_SERVER_RESTART. It's make the difference on the performance of your environment. Check the size of your physical log DBSpace with that changes using onstat -g ckp. If after the changes you have checkpoints triggered by Plog, you must add space on the phisical log DBSPace. I noticed that the values of lru_min_dirty and lru_max dirty are too little. Try to set LRU_MAX_DIRTY=85 and LRU_MIN_DIRTY=75, and let your instance adjust this watermarks according his workload. Increase the value of LOCKS of your instance (if you have free available memory), because I noticed that you had lokwaits. The value of each lock is 44 bytes Try this: Configuration Parameters: AUTO_CKPTS 1 RTO_SERVER_RESTART 1800 BUFFERPOOL size=2K,buffers=2400000,lrus=240,lru_min_dirty=85.00000,lru_max_dirty=75 .00000 AUTO_LRU_TUNING 1 I think if you don't have storage bottlenecks (Storage I/O in 500MB/sec read/write, at least), the performance gains would be good. Regards, Alberto. De: informix-list-bounces@iiug.org [mailto:informix-list-bounces@iiug.org] Em nome de Habichtsberg, Reinhard Enviada em: quarta-feira, 13 de julho de 2011 09:29 Para: informix-list@iiug.org Assunto: Re: AIO and Buffer Waits Hi Art, BTR is 48,57 per hour. According to Lester Knutsen should be < 6 per hour. BUFFERPOOL looks like that and AUTO:LRU_TUNING is turn on: BUFFERPOOL size=2K,buffers=2400000,lrus=60,lru_min_dirty=0.200000,lru_max_dirty=0.400000 AUTO_LRU_TUNING 1 onstat -p (last onstat -z 14 hours ago) IBM Informix Dynamic Server Version 11.50.FC7W3 -- On-Line -- Up 54 days 01:54:20 -- 8984576 Kbytes Profile dskreads pagreads bufreads %cached dskwrits pagwrits bufwrits %cached 142604256 1410570080 59284431722 99.76 61872253 98433717 227436232 72.80 isamtot open start read write rewrite delete commit rollbk 52046254582 216281491 457585123 48750470513 17730429 30711047 19268677 18995943 1446067 gp_read gp_write gp_rewrt gp_del gp_alloc gp_free gp_curs 14 0 9 0 0 0 0 ovlock ovuserthread ovbuff usercpu syscpu numckpts flushes 0 0 0 317120.96 38441.46 134 163 bufwaits lokwaits lockreqs deadlks dltouts ckpwaits compress seqscans 20157325 126775 1927754927 0 0 5149 6933881 1205132 ixda-RA idx-RA da-RA RA-pgsused lchwaits 66150442 1636318 30850997 97300297 20233621 What would be a reasonable value for lrus? TIA, Reinhard. From: Art Kagel [mailto:art.kagel@gmail.com] Sent: Wednesday, July 13, 2011 12:24 PM To: Habichtsberg, Reinhard Cc: informix-list@iiug.org Subject: Re: AIO and Buffer Waits Most bufwaits are caused by contention for the LRU queues, so increase the LRU queues for your buffer caches to maximum first. Use the BTR calculation and observing activity in onstat -P over time to determine if you may also need more buffers. Art Art S. Kagel Advanced DataTools (www.advancedatatools.com) Blog: http://informix-myview.blogspot.com/ Disclaimer: Please keep in mind that my own opinions are my own opinions and do not reflect on my employer, Advanced DataTools, the IIUG, nor any other organization with which I am associated either explicitly, implicitly, or by inference. Neither do those opinions reflect those of other individuals affiliated with any entity with which I am affiliated nor those of the entities themselves. On Wed, Jul 13, 2011 at 2:49 AM, Habichtsberg, Reinhard <RHabichtsberg@arz-emmendingen.de> wrote: We see the following waitstates on one of our Informix 11.50FC7W3 servers. He is running OLTP but also Decision Support. (sum) 81125424549,00 reason aio (sum) 62105207794,00 reason buffer (sum) 61713678094,10 reason mt yield n (sum) 61561288458,60 reason mt yield (sum) 29339327941,19 reason mt ready (sum) 15649570039,20 reason running (sum) 3340893922,100 reason log i/o (sum) 1550362515,000 reason mt yield 0 (sum) 1439244418,300 reason checkpoint (sum) 162272243,0000 reason lock (sum) 157356745,3000 reason mutex (sum) 21639138,60000 reason full Q What do you suggest to reduce IO and Bufferwaits. My idea is to increase the buffers. TIA, Reinhard. _______________________________________________ Informix-list mailing list Informix-list@iiug.org http://www.iiug.org/mailman/listinfo/informix-list