Re: Keeping clean pages in shared memory

> > 1) If a buffer is written to, say, every 30 seconds and you flush it > once a minute, you'll do five times as many disk writes for it than > with a 5-minute checkpoint That's what we thought but it wasn't the case. We basically came to the conclusion that the page cleaning activity was done when the disk was not busy. After all it requires 30ms at most to write a page away. The fact that you are doing five times the number of disk writes is only an issue if the disk is already highly utilised. > 3) A nice big checkpoint every five minutes, with sorted writes, is > likely to have a lower head-movement-to-pages-written ratio than a > small fuzzy flush every minute. Big checkpoint are noticeable by the user which is why we steer clear of them. > Couple of grey areas though. > - Are critical sections suspended during a fuzzy flush? (If they are > then my head contention argument is less strong) I don't think they need to be. > - I can't recall whether fuzzy flushes do sorted writes. I seem to > remember reading one manual or document that said they do, another that > said they don't. I can't work out how it knows since it does the least recently used page first. David W Chan Objective Research Ltd dwc@cix.compulink.co.uk