Hi, sry, when I said maximum_object_size I wanted to say
maximum_object_size_in_memory :P.
Right now (i don't know how) it's working. Here is some info:
Cache information for squid:
Request Hit Ratios: 5min: 92.0%, 60min: 92.3%
Byte Hit Ratios: 5min: 35.5%, 60min: 34.7%
Request Memory Hit Ratios: 5min: 61.6%, 60min: 60.2%
Request Disk Hit Ratios: 5min: 0.5%, 60min: 0.4%
Memory usage for squid via mallinfo():
Total space in arena: 79176 KB
Ordinary blocks: 71934 KB 9224 blks
Small blocks: 0 KB 0 blks
Holding blocks: 432 KB 2 blks
Free Small blocks: 0 KB
Free Ordinary blocks: 7241 KB
Total in use: 72366 KB 91%
Total free: 7241 KB 9%
Total size: 79608 KB
It's very VERY good compared with I was getting before.
My box is a P4 3.0 with 4 GB RAM under FedoraCore 5.
Cheers and thanks,
Santiago del Castillo
K Kadow wrote:
On 11/1/06, Santiago del Castillo <santiago.del.castillo@xxxxxxxxxxx>
wrote:
Hi, i'm having problems with MEM_HITs, the mem_cache is 2GB, the
maximum_object_size is 4096 KB and i'm having just 10.7 % MEM_HIT ratio.
What is your maximum_object_size_in_memory?
The default is very low.
I've been told here that, depending on your host OS and how much RAM
you have, you can actually increase hit ratios and performance by
setting both cache_mem and maximum_object_size_in_memory very low, and
relying on the operating system to intelligently use memory to cache
disk blocks and serve disk hit answers out of the OS cache.
This assumes the server is dedicated to Squid and that the OS is
smarter about caching frequently used objects in memory than Squid is.
Kevin
