Re: chekcpoints, BUFFER settings, and optimization

In article <348F1333.4F79@bloomberg.com>, "Art S. Kagel" <kagel@bloomberg.com> writes >David Williams wrote: >> >> In article <66madi$osk@cssun.mathcs.emory.edu>, Serafim Fonseca >> <inf@dpx20.growela.pt> writes > >> >} In article <881611278.1353908229@dejanews.com>, >> >} jneugebauer@ameritrade.com writes >> >[...] > >> >David Williams writes > >> >[...] >> >} ># Shared Memory Parameters > >> >} >LOCKS 100000 # Maximum number of locks BUFFERS 5000 # Maximum number > >> >} BUFFERS 5000!! >> >} Run onstat -b and look at the last line it will end in page size >> >} either 2048 bytes or 4096 bytes (2 or 4K). > >> >} I would up BUFFERS by 100Mb and run sar -p 1 1 / sar -q 1 1 >> >} make sure you are not getting paging/swapping. Keep on increasing >> >} by 100Mb unti paging starts then set it to the highest value which >> >} has no paging. > >> >} Run onstat -g ath. Are there kaio threads running. If so set >> >} NUMAIOVPS=1. Otherwise increase NUMAIOVPS to the number of disks you >> >} have (up to 32). > >> >} Set cleaners = 6 (same as NUMCPUVPS). Page cleaner threads only queue >> >} i/o (pass it to kaio threads or AIO VPs) they do not actually do the >> >} I/O. > >> >Is this a rule? To have the same number of cleaners as the number of >> >NUMCPUVPS. >> >My machine has 1 NUMCPUVPS and I've CLEANERS=8 should I change that? > >> Yes, cleaners threads run on cpu vps. Hence with 1 CPU VP only 1 can >> be active at any one time. More than 1 is a waste of time. > David Williams > >Dave I have to disagree. I agree that additional cleaners are of >limited use during normal LRU writes but they are needed for fastest >checkpointing. Since at checkpoint time each chunk is assigned to >another cleaner thread until the threads are exhausted, and since that >one thread, as has already been pointed out, is only scheduling the >actual I/Os with either the AIO VPs or KAIO VPs, that thread will block >on each I/O that it schedules and single thread writes to your disks. Are you sure it blocks, I though it was Asynchronous I/O... >You must have multiple cleaners, even with a single CPU VP, since the >other cleaners can wake and schedule the I/Os that they are responsible >for while the first cleaner is blocked waiting for I/O service. > >Because of the asynchronous I/O of ODS the old OL5.xx recommendation of >one cleaner per LRU is overkill. But one cleaner for each 2-3 LRU >queues is probably just about right, no matter how many or how few CPU >VPs you have available. This will ensure that your checkpoints are not >bottlenecked by single threaded I/O. > >Art S. Kagel -- David Williams