On Wed, 2005-01-05 at 11:10, Matus UHLAR - fantomas wrote: > On 28.12 16:36, Paul Clayton wrote: > > We have squid Version 2.5.STABLE7 installed, Redhat Fedora Core 1. 1Gb > > Ram Pentium Xeon 3.06Ghz > > > > Our cache size is 7.5Gb for 120 users. Average daily downloads vary > > between 2Gb-6Gb. During some testing, we noticed that http downloads > > were significantly slower than bypassing the cache and going direct. > > > > Typical readings were 19Kbytes per second through the cache versus > > 250Kbytes per second going direct. Playing with the memory did make > > impact, but the optimum we can attain is having a cache memory of 256Mb > > and 16Mb memory pools. ALthough messing with the memory pools, did not > > make much difference. > > I would say this problem is outta squid. It may be wrong network setup > (half duplex vs full duplex), not using DMA on disk (is it IDE disk?) > and probably not ideal filesystem. > > I'd recomend using bigger disk cache - for accomodating week's traffic you > should have 30GB cache imho. Tuning up filesystem may help too, using xfs > or reisersfs (with notail option) on cache disk and probably use dedicated > cache drive.
Nope... The book - "Squid - the definitive guide" shows that ext3 is still faster then reiserfs. Search the archives for the bit which I wrote and quoted from the book -- Ow Mun Heng Gentoo/Linux on DELL D600 1.4Ghz 98% Microsoft(tm) Free!! Neuromancer 14:44:13 up 5:42, 6 users, load average: 0.48, 0.59, 0.51
