Re: Reducing checkpoint times on cooked db

The database is embedded in a product and by default will use cooked files because of configuration issues with setting up raw files. It will be an option to set up to use raw, but we want the performance of the cooked files to be as good as it can. I thought it will require additional writes to the database because the dirty buffers will get written by the cleaners more often and you will not get the advantage of multiple writes to the buffer before it is put onto disk. Neil Truby wrote: > > BART FISHER wrote: > > > > We are running Online 7.2 and encountering 45 second > > checkpoints when the db files are cooked. When using > > raw files the checkpoints go down to 5 seconds. We > > are heavily doing insert/delete ops (100 per second) on > > 2 tables. Changing the LRU_MIN and LRU_MAX to very low > > values helps a little (down to 40 secs) but our disk > > is smoking and the overall ops per second suffer > > because of the additional disk writing. > > "Additional"? Or "slower"? > Your question begs another: why not revert to raw disk? But if you move > from raw to cooked I suppose you have to look at things like the number > of CLEANERS and AIOVPS (normally just one when using KAIO). Presumably > iostat can help you identify where your poor service times are occurring > - perhaps you could try some I/O balancing? > > Neil Truby > Londis Holdings > London, UK