Felix New wrote:
Hi,all:
Our squid box's performance is not very good(40000 reqs/min), cpu
load is very high, and the traffic can't great bigger than
100MBits/s.(Gbits/s network).
We use aufs scheme now.We have a test for coss, and the result is
not very good also(cpu overload also).
can we optimize continue?Is the best performace that our hardware
can achived?where is the bottle-neck?
what is the best performance of your box?
there are some infomations:
++++++ hardware:
cpu: Xeon 1.8G(Hyper-Threading)
os: RHAS 4.3
file discriptor num: 32768
mem: 2G
disk: OS: 17G(scsi) for OS,
cache:73G(SCSI,10000rpm for cache data)
Only one spindle for cache? That could hamper performance, but I don't
see much in the way of IO wait.
++++++ squid.conf
about 100 ACLs
What kind of ACLs? I imagine that you are not using any regex, but
figure it's prudent to ask.
cache_mem 1500 MB
cache_dir coss /dev/sdb1 50000 max-size=10000000 block-size=8192
(some server use aufs: cache_dir aufs /cache/cache 50000 240 256,
and sdb1 is mounted with options of async and noatime)
Ouch! 2 GB total RAM, and you are using 1500MB for cache_mem with a
50GB cache_dir? According to the calculations from the FAQ
(http://wiki.squid-cache.org/SquidFaq/SquidMemory), that is going to
lead to at least 2GB of RAM used by Squid ALONE. That does not account
for any memory used by the OS. It would seem that your disk cache is
not heavily populated...
++++++ top:
top - 11:37:46 up 42 days, 12:00, 1 user, load average: 1.00, 0.99,
0.91
Tasks: 54 total, 3 running, 51 sleeping, 0 stopped, 0 zombie
Cpu(s): 41.8% us, 8.3% sy, 0.0% ni, 47.9% id, 1.3% wa, 0.7% hi,
0.0% si
Mem: 2074980k total, 1979160k used, 95820k free, 686568k buffers
Swap: 4192956k total, 3404k used, 4189552k free, 355632k cached
Chris
