On 09/30/2010 04:27 AM, Mr. Issa(*) wrote:
Dear All,
Iam facing a server load of 2.0 almost all the time and a whole of
32GB of rams are consumed on the server used for squid production
My question is Why the 32GB of rams are consumed and why the server is
LOADING all the time
The squid is used in transparent mode, and cache_replacement_policy heap LFUDA
memory_replacement_policy heap GDSF
client_http.hits = 45.562854/sec <<<
Based on that, if we calculate in TCP, overhead, MySQL,
Apache/NGINX/Lighttpd (especially lighttpd which has known memory leak
in the 1x branch which isn't going to be addressed till 2x branch) that
seems about right for an unoptimized server that isn't designed around
the application but designed by a System Admin who builds generalized
servers.
Do a ps auxf and show us what memory it's actually using according to
the system. Whoever designed your system also improperly designed it's
SWAP, normally for a consumer system you need twice the SWAP you do
memory, on a production server you should have at least 1/3 the memory
as SWAP you barely have 1/31, this can increase memory usage because the
system won't push to the SWAP when switching in and out of memory. SWAP
is also a common misunderstood subject by system admins.
"A server load of 2" is incorrect and you misunderstand what the 2.0
means, that is the average number of processes awaiting resources on
your server, this can include i/o, other processes and CPU time. Each
number is representative of a time frame so: 2, 2.5, 2.3 is 2 processes
every 1 minute, 2.5 processes every 5 minutes and 2.3 processes every 15
minutes. You need to be looking at CPU utilization. And just to give
you a heads up on the number, that actually seems perfectly low for a
server that might be pretty busy.
