Hi everyone, I'm running varnish (2.0.2) on Linux x64 inside Virtuozzo Containers 4.0. The container has got 2GB core memory assigned.
I have a backend that holds about 30GB binary objects. Varnish is set to assign a ttl of 30 days to each object it caches. I start varnish with -s file,/var/lib/varnish/varnish_storage.bin,50GB because I assumed it would be wise to have a cache file that is larger than the max. amount of backend objects. However I see a sawtooth behaviour in cache hit rate that corresponds with the server's virtual memory allocation. Munin stats: http://www.flickr.com/photos/25013...@n02/sets/72157618375799306/ It looks like varnish hits the container's memory limit, purges all memory and all objects from the store and starts from the beginning. If I set -s file to 1500MB virtual memory usage is at a constant 1.5GB as one would expect. However my goal is to have all objects (30GB) on the varnish server's local disk, and the most requested 1.5-2GB in it's core memory. I thought that setting the file store size to 50GB would do the trick and the server than would decide which objects are hold in memory and which are read from local disk, however the 100% MISS ratio whenever the server hit's it's virtual memory allocation limit tells me I've done something wrong. Any ideas as to what that might be? Am I using varnish the wrong way? -- Lukas _______________________________________________ varnish-misc mailing list [email protected] http://projects.linpro.no/mailman/listinfo/varnish-misc
