Re: Ree: long checkpoints

tomL wrote: > thanks for the reply. > I am curious why I see such low numbers on MIN/MAX, with many > people setting them to 1/2 and apparently 9.4 giving the Admin the > ability to even go to a decimal setting! (We are going to try lowering > ours to 10/15 to see if that will make enough of a difference - I am > hoping that is enough to see a change.) Is the idea to fill up very > little and have the checkpoint be very short? Can someone explain > further? > > "Mark Denham" <mkdenham@attbi.com> wrote: >>Lower your LRU min and max for starters. That will reduce the volume of >>dirty pages that have to be written out as part of the checkpoint process. I >>would suggest you try 4/2 as a starting point. At a checkpoint, most (but not quite all; think of fuzzy checkpoints - or look them up in the manuals) dirty pages have to be written out to disk. Suppose you have 1 GB of shared memory. 1% of 1 GB is 10 MB (approx). It can take a while to write 10 MB to disk - which increases the length of your check point. Suppose you have a big machine - maybe 100 GB of shared memory... By allowing LRU_MIN_DIRTY and LRU_MAX_DIRTY to be set to fractions of a percent, you can reduce the amount of data to be written at the check point - speeding them up. There are other things you can do to reduce your checkpoint times; a look for checkpoint time in Google should be instructive. -- Jonathan Leffler #include <disclaimer.h> Email: jleffler@earthlink.net, jleffler@us.ibm.com Guardian of DBD::Informix v2003.04 -- http://dbi.perl.org/