Michael S. Fischer
michael at dynamine.net
Mon Nov 10 15:11:46 EST 2003
Gregory Bond wrote: > cs1spw at bath.ac.uk said: > >>Actually we do - every 5 minutes we check the timestamp to see if an >>update has been made. We're trying to avoid making the call to the >>filesystem to check the timestamp on every request. > > > Actually, my gut feeling is that a stat() on a file that is still in the cache > is not going to be noticeably more expensive than the gettimeofday() call you > need to do to decide whether five minutes are up. So give it a whirl with the > stat() on every call and see what that does to your performance, you might be > surprised. Poster is correct; if your machine's inode cache is too small, you can adjust it via sysctl kern.maxvnodes (BSD); Linux's inode cache is dynamic. --Michael
|