Chris Robertson wrote:
Roger 'Rocky' Vetterberg wrote:
Hi list.
[snip]
I have tried to configure them according to what I read in
documentation and FAQ's, but I run into heavy swapping, "Unable to
allocate" errors or just bad performance. It seems I'm having
problems finding a good balance between performance and stability.
Have you read through the genuine Squid FAQ section on memory
(http://wiki.squid-cache.org/SquidFaq/SquidMemory)? It's probably a
good place to start.
Yes, several times. It explains in great detail what certain errors mean
and how to see how much memory is used for what, but does not give much
help when it comes to calculating what values to use.
Could someone give me some rough figures to use for cache_mem,
cache_dir, L1, L2, Q1 and Q2?
Personally, I leave cache_mem at the default value, and trust my OS to
cache disk accesses. As for the cache_dir, the consensus seems to be
not to fill your partition beyond 60% for best performance.
Would I benefit from using diskd, or should I run with normal UFS?
Use diskd or aufs.
Both servers are armed with hardware raid on 15k drives, so the disk
I/O should be pretty decent.
Just as long as you are aware that RAID 0 really doesn't have the "R",
and as far as I am aware, most RAID controllers only show significant
performance increase when calculating parity (RAID 3-6), you might be
better off using one spindle for the OS and creating separate cache_dirs
on each of the other spindles. That way a SPECIFIC disk has to die for
your proxy server to go down.
Im not really worried about a server going down. The reason I have two
servers is to be fully redundant. Im planning on implementing CARP and a
round-robin DNS entry as soon as I have both servers tweaked.
--
R
