Re: Minimum advisable for LRU_MIN_DIRTY

Omar Muñoz wrote: > Hi. > > I deal with an IDS 9.40 FC7 under Solaris 9 > transactional installation with LRU_MAX_DIRTY = 2 and > LRU_MON_DIRTY = 1, and 950000 in BUFFERS. Even when > cleaners starts very soon, my checkpoints use to be as > big as 50 seconds during peak hours (I've tried to > increase or decrease checkpoint interval with no > success, nowadays I assign 4 minutes). I'm wondering > if setting LRU_*_DIRTY to 1 and 0 would work, or if > it's possible to set values as 1.5 and so on. Let me > add that the working environment is OLTP, but a lot or > writes may happen very often, operations that write in > average 50 Mb on logical logs in 4 minutes. Is my > approach correct, or should I try a different point of > view? > > Thanks in advance. > Omar, First, yes, in 9.40 and later releases you can set the LRU_MIN/MAX_DIRTY to fractional values. You typically want to have under 2000 buffers to flush at checkpoint time in an OLTP system, and since on average the LRUS will be about half way between those two values, you want to set so that LRU_MAX_DIRTY equals about 3500 buffers or 0.35% with LRU_MIN_DIRTY set to 0.10%. Or you can set them to 0.4 and 0.0 respectively, but there have sometimes been problems with setting LRU_MIN_DIRTY to zero, so I tend to avoid that. Also, since you are running 9.40, you should keep fuzzy checkpoints disabled. Fuzzy checkpoints were a misguided attempt to reduce checkpoint times for larger servers like yours, but it doesn't live up to its hype. Keep in mind however, that the number of dirty buffers at checkpoint time is only one component of longer checkpoint duration. The checkpoint cannot begin while any session is in a critical section of code within the engine. Larger transactions will tend to keep sessions in critical sections longer and large numbers of sessions will increase the probability that the checkpoint will have to wait for one or more session to exit its critical section before beginning. This wait time is included in the checkpoint duration calculation because all existing sessions are blocked from entering critical sections once the checkpoint is initiated. So, look also to reducing transaction sizes and reducing the number of client connections if possible (rarely achievable I know). Art S. Kagel Oninit > Omar Muñoz > > > =========================================================================================== Please access the attached hyperlink for an important electronic communications disclaimer: http://www.oninit.com/home/disclaimer.php ===========================================================================================