On Mon, Dec 04, 2006, Nguyen, Khanh, INFOT wrote: > Hi, > > I have a question about cache_mem, hope that someone has some insight > about it. > > I run squid 2.6 on AS linux 4, update 3. My server has 8 GB memory, 20 > GB virtual memory, 50 GB hard disk for squid cache. If the OS is paging, either because the system is running low on memory or the OS is paging stuff out 'in case' it needs to allocate some memory at a future time then this'll certainly happen. Try some profiling and watch the output of 'vmstat 1' to see whether it does page. > Is it true that allocating large cache_mem would decrease the > performance of the cache server? For example, I was told that with 3 GB, > load is 400 r/s, with 2 GB, load is 1200 r/s, object size is 1-100KB. > That is 3 times difference. It goes against common sense that memory > would help performance, not hurt. Maybe cache_mem works differently and > the more it has, the worst the squid performs. Or something in the load > generator which makes it worst. In which condition, more memory will be > worst? What would be the best number for cache_mem? It shouldn't be performing any worse with more cache_mem. I've certainly not seen any measurable performance difference between 512meg and 1 gigabyte of cache_mem. I'm unable to test this assumption as I just don't have servers with that much RAM available (and if anyone would like to donate to the Squid project so I could have such a server available then I'd be very grateful and also be able to post actual performance figures on the Wiki..) There's certainly an issue with CPU usage with larger objects held in memory but only if you try caching large objects - say, above a megabyte. Adrian -- - Xenion - http://www.xenion.com.au/ - Hosting and Commercial Squid Support -
