Also sprach Henrik Nordstrom <hno@xxxxxxxxxxxxxxx> (Sat, 14 Jan 2006 14:26:20 +0100 (CET)): > On Sat, 14 Jan 2006, Richard Mittendorfer wrote: > >> Why I ask is because diskd is known to be somewhat slow on large > >cache > > > > Not really large. 2x 1G. It's no storage bottleneck I believe. > > large cache hits == hits on largeish cached objects. Oh, sure. Didn't had enough coffee this morning.. :-) > >> hits in certain situations UNLESS there is sufficient traffic to > >keep > Squid reasonably buzy (i.e. problems if you are the only user, > >or very > few users). And the same for aufs in older versions of > >Squid. > > > > See. Would fit. > > A quick test if this is your problem is to reconfigure your Squid to > use the ufs cache_dir type. 7,30M/s. That helps. Little bit slower with aufs: 6,85M/s. hmm.. However, aufs (posix-threads?) seem to like/malloc a lot of Memory. Running on mere 256M Ram and offering a good many services, Commited_AS klimbs to 550M (340M w/ diskd). And Squid hasn't been used yet. I suppose it will get swapped out way more easily. Will consumed memory be much higher with aufs in contrast to diskd(/ufs)? I'll see in a few hours/days. > Regards > Henrik THX ritch
