Checkpoints
I have seen recommendations that setting LRU's 128 can, among other changes, help to reduce long checkpoints of 10, 15, 20 or more seconds. For the most part this recommendation seem to be for 7.3x versions of Informix. Is there a problem with using the same setting for 7.24 versions? I seem to remember reading something about a bug with high LRU settings. Has anyone used max setting on LRU's for a 7.24 engine without encounter a problem? Sent via Deja.com http://www.deja.com/ Before you buy.
We used a 7.24 engine on AIX with LRU's at 127 with no problems, we also reduced the time between checkpoints. The most I've seen (once) was a 3 second checkpoint. We do them every 60 seconds. Subsequent to that we have upgraded to 7.31 with the same settings across the board. brett In article <8opjcb$jh$1@nnrp1.deja.com>, dechase@my-deja.com says... > I have seen recommendations that setting LRU's 128 can, among other > changes, help to reduce long checkpoints of 10, 15, 20 or more seconds. > For the most part this recommendation seem to be for 7.3x versions of > Informix. Is there a problem with using the same setting for 7.24 > versions? I seem to remember reading something about a bug with high > LRU settings. Has anyone used max setting on LRU's for a 7.24 engine > without encounter a problem?
Use 127 instead. There is a harmless but annoying bug in 7.24 at LRUS=128 which will hound you about having a zero length PHYSICAL and LOGICAL log buffer size set. Using 127 gets all the benefit without the pain. BTW setting CLEANERS also to 127 is even more important to checkpoint times than LRUS. Art S. Kagel dechase@my-deja.com wrote: > > I have seen recommendations that setting LRU's 128 can, among other > changes, help to reduce long checkpoints of 10, 15, 20 or more seconds. > For the most part this recommendation seem to be for 7.3x versions of > Informix. Is there a problem with using the same setting for 7.24 > versions? I seem to remember reading something about a bug with high > LRU settings. Has anyone used max setting on LRU's for a 7.24 engine > without encounter a problem? > > Sent via Deja.com http://www.deja.com/ > Before you buy.
Well we took a chance before the replies arrived and set CLEANERS & LRU's to 128. Indeed, we did get the "WARNING! Physical Log size 0 is too small..." and the "WARNING! Logical log layout may cause OnLine to get into..." messages. Not knowing these were harmless we stopped the engine and rechecked ONCONFIG. It looked fine. We shrugged and started the engine back up. Informix crashed with a fatal shared memory error and locked up the OS. Even console commands were ignored with a similiar "not enough shared memory" type error message. We were unable to reboot so we powered down and back up again. Informix came up with the same two warnings. We shutdown Informix, changed CLEANERS & LRUs back the previous settings. Informix came up normal. The online.log contained the following message logged at the time of the crash: shmget: [ENOMEM][12]: key 52574802: out of shared memory, check system SHMALL 23:43:29 mt_shm_init: can't create virtual segment" Now we're a just a bit timid about changing CLEANERS & LRU's again. DChase In article <39B51CDF.4DF96E6E@bloomberg.net>, kagel@bloomberg.net wrote: > Use 127 instead. There is a harmless but annoying bug in 7.24 at LRUS=128 > which will hound you about having a zero length PHYSICAL and LOGICAL log > buffer size set. Using 127 gets all the benefit without the pain. BTW > setting CLEANERS also to 127 is even more important to checkpoint times than > LRUS. > > Art S. Kagel > > dechase@my-deja.com wrote: > > > > I have seen recommendations that setting LRU's 128 can, among other > > changes, help to reduce long checkpoints of 10, 15, 20 or more seconds. > > For the most part this recommendation seem to be for 7.3x versions of > > Informix. Is there a problem with using the same setting for 7.24 > > versions? I seem to remember reading something about a bug with high > > LRU settings. Has anyone used max setting on LRU's for a 7.24 engine > > without encounter a problem? > > > > Sent via Deja.com http://www.deja.com/ > > Before you buy. > Sent via Deja.com http://www.deja.com/ Before you buy.