Re: Informix tuning

"Mark D. Stock" <mdstock@informix.com> offerred: +Having said all that, there is nothing wrong with forcing a checkpoint +every minute, but then you must expect your write cache figures to +suffer accordingly. Considering the current (and long time) checkpoint behavior of blocking transactions while they occur, I would suggest that forcing a checkpoint every minute is generally a bad idea (frequent checkpointing can be the kiss of death for high throughput oltp applications). The only real advantage of such frequent checkpointing is the minimizing of fast recovery time should a system crash occur. Much more palatable would be to keep the LRU_MAX_DIRTY and LRU_MIN_DIRTY low and let lru writes keep the buffer pool reasonably clean in between checkpoints. I steadfastly maintain that write cache stats should be ignored in almost all cases. Why? Because, in fact, ALL writes are cached. Your effective write cache ratio is always 100%. What the stat really measures is the ratio of bufwrites to diskwrites. Since writes are almost always asynchronous to database processing (with v7, all are effectively async, even fg writes, though the latter will always cause the sqlexec thread to yield), who cares how much i/o is taking place between checkpoints as long as it is not hurting overall system performance. Much more applicable are the i/o queue lengths, which tell you if specific devices are realizing i/o bottlenecks. For cacheing, stick to worrying about read cacheing (which does directly affect your app performance) and forget about write cacheing. Dave ** Dave Kosenko <davek@summitdata.com> ** Director of Training Services (732) 469-4070 ** Summit Data Group (an Informix Authorized Education Center) ** Find my advice useful? Let me teach you everything I know about ** Informix. Sign up for OFFICIAL Informix training at SDG. ** For details, see http://www.summitdata.com/training