Re: Checkpoint takes long

Kai Shen wrote: > on our SCO 5.0.4 machine we are running ODS 7.20 UC3. During the day > when appr. 60-80 persons are using the machine, we permanently > experience checkpoints that take 10 seconds or longer. I tried > changing the time between checkpoints from 1 minute to 10 minutes > and back to the default of 5 minutes, the checkpoints will always > take very long. We have set the physical log size to 20000 and use > 30 logical logs, each 2048. Our buffers are set to 100000, LOCKS > 300000, LRU_MAX_DIRTY 2, LRU_MIN_DIRTY 0, LRUS 12. > Which parameters affect the checkpoint time mostly? Number of dirty buffers, which corresponds to BUFFERS and LRU_M??_DIRTY, plus the number of threads available to do the writing to disks, and the number of disk drives, and whether you're using some obscure RAID configuration which takes forever to write, and ... Since you've got your LRU_MAX_DIRTY set a 2%, you could have 2000 pages to write at a check point -- 4 MB of data with a 2 kB page size. You could decrease that by setting LRU_MAX_DIRTY to 1, I suppose, but that's the end of the road, I think. I've never heard of anyone who used LRU_MAX_DIRTY = LRU_MIN_DIRTY = 0, though I suppose it would work, nominally with instant write-to-disk for dirty pages. In terms of simple engine tuning, you need to examine the page cleaning resources available. The other parameters are either not material or close to optimal. The rest of the analysis should be spent on the disk i/o system -- how is it laid out, and what could speed it up. -- Jonathan Leffler (jleffler@informix.com, jleffler@earthlink.net) Guardian of DBD::Informix v0.60 -- see http://www.perl.com/CPAN #include <disclaimer.h>