Hey Chris, Could you comment on the "Internal Marker Objects", as in, are they stored just in COSS or elsewhere too? Thanx. On Thu, 2010-01-07 at 10:21 -0800, Chris Woodfield wrote: > One thing we've noticed is that until the COSS store is rebuilt, no objects get saved to it (each attempt shows up as SO_FAIL in cache.log). Complicating this further is the fact that -F doesn't seem to work for COSS rebuilds - I know I opened a bug on this but I can't seem to find it at the moment... > > -C > > On Jan 6, 2010, at 5:18 PM, Manjusha Maddala wrote: > > > > > On a related note, > > > > if Squid is configured to use AUFS and COSS for its cache_dirs, does it > > store the "Vary Internal Marker Objects" only in the COSS file or > > somewhere else too? For my setup, I notice the marker objects which > > provide "vary" meta-data for all the cached pages are stored in COSS. > > > > If there is a COSS cache_dir, Squid attempts to store all the cached > > objects less than the max_size for COSS object in the COSS file. As a > > result, some of the "Internal Marker Objects" get evicted out (during > > COSS recycle) thereby resulting in low cache hit ratio for my setup. > > > > > > > > On Wed, 2010-01-06 at 16:58 -0800, Chris Robertson wrote: > >> Jeronimo Garcia wrote: > >>> Hi all. > >>> > >>> I've been looking at a bunch of benchmarks about storage schemes for > >>> squid and coss looks > >>> rather impressive , but , reading squid the definitive guide i got to > >>> know that the code might be a bit beta/unstable. > >>> > >> > >> http://wiki.squid-cache.org/Features/CyclicObjectStorageSystem > >> > >>> Now i didn't really check when this book was written but I've been > >>> coming across with it for a bunch of years already so i imagine is not > >>> very new. > >>> > >>> Does some of you have some real experiences with coss? > >> > >> Squid 2.7STABLE6 > >> > >> My COSS dirs are about 45 GB each on bare partitions (cache_dir coss > >> /dev/sdc1 46080 max-size=51200 max-stripe-waste=32768 block-size=4096). > >> Each server passes about 100 GB of traffic per day (to my customers, > >> ~25% from cache), with a peak requests/second of around 150. No issues > >> with stability (currently I have over 50 days of uptime). > >> > >>> I'm also very > >>> interested on squid's start-up time when using coss. > >>> > >> > >> Squid starts up just fine but thrashes the disks hard for about half an > >> hour while it reads the COSS dirs to build the index. There is no > >> noticeable effect on performance. > >> > >>> Many Thanks > >>> -J > >>> > >> > >> Chris > >>
