======= 2006-05-15 12:39:52 Adrian Chadd wrote:======= > >On Mon, May 15, 2006, joe wrote: > >> >COSS is currently not stable in Squid-2.5STABLE13. >> >> I know COSS is not stable, but the large throughput is attract me. > >Yup, I know its attractive. > >> >COSS is also not very good at storing objects that big; the default stripe >> >size is 1 megabyte and I haven't tested it with larger stripe sizes. >> >COSS is also much better at storing smaller objects, under 64k. You should >> >configure an aufs or diskd cache-dir after the two COSS stores to catch the >> >objects that COSS can't cache. I'd suggest using diskd under FreeBSD 4.10. >> >> I have not use big objects. I use squid to some WEB service and my mostely objects size less than 100k. >> I set the stripe size with "--with-coss-membuf-size=2097152" when I configure the squid source code. Is that the reason of the WARNING message "got unused STORE_META type -72" appear? > >I haven't tried it with a membuf stripe size besides the default. > >The warnings are because of race conditions which I've fixed and I'm now testing. > >> Before use COSS, I had use the DISKD an UFS Storage Scheme under FreeBSD 4.10.But I barge up against too. Sometimes, the "cache_dir"(/home) subarea is full when I see with "df" command. But if I see with "du" (du -hs /home) , I found the subarea use less than 50%. i just wonder whether the bugs occurred by the squid or the FREEBSD. > >Weird! > >> >I'm getting some strange stale object errors when testing with Polygraph; >> >I haven't investigated enough yet to see why they're happening. >> >I'm not running COSS locally yet either: I've spent the last couple of weeks >> >ironing out the exact bugs you've been seeing. >> > >> > >> Thank you. I will look forwards and try it. > >I'll email the list when the code is ready for testing. It shouldn't be that long now. > > > > >Adrian > > = = = = = = = = = = = = = = = = = = = = Thank you very much! Joe Cen joecen@xxxxxxx 2006-05-16
