Re: Batch tuning...

dday@rheem.com wrote: > > I recently had an "independent" contractor come in and analyze a check > point problem we were having (in excess of 5 minutes sometimes) and > have implemented a new onconfig with his suggestions. The oltp has > really improved. Our problem now is that batch processing, both imbedded > cobol and 4ge's, has gone to hell (really sloooow). Have we gone too far > in our tuning attempt? Any suggestions? > Thanks. David Day I agree with the other responses. > # BUFFERS 150000 # Maximum number of shared buffers > BUFFERS 5000 # Maximum number of shared buffers Hmmm. 4 gig of ram, buffers reduced from 300Mb to 10Mb. IMHO reducing buffers to deal with a checkpoint interval problem is an absolute last resort. As others have pointed out, there are other ways to deal with the problem that should work. If your I/O subsystem is really stressed, you can even set LRU_MIN_DIRTY/LRU_MAX_DIRTY as low as 1/2 to force almost all writes as LRU writes. Under these settings if you are still getting large dirty buffers at long checkpoints then you need a new I/O subsystem (means that the LRU writes are not keeping up). If you have the option of moving to IDS 2000, you will find "Fuzzy Checkpoints" a good way of reducing buffer flushing at (most) checkpoints. Did you get a refund? Brett Randall <snip>