On Tue, 22 Mar 2011 20:30:11 +0100, Dejan Zivanic wrote:
Regards,
we have heavy load (over 6k requests per minute) intercepting squid
loading about 70-80Mbps traffic.
I have notices that CPU usage of squid process never goes down from
50% and usually goes up to over 90%.
We plan to upgrade to 120Mbps link and this can be major problem if
we cannot solve it.
Every suggestions will be appreciated...
Believe it or not this is pretty good for Squid-3. We usually see
people having performance problems from around 50Mbps. Though your 6k
RPS is lower than those testers used.
For scaling out once squid hits its limit you will need more than one
edge Squid.
This is what comes to mind from me. They are just things to look at,
some may help some may not, so testing is worthwhile first:
* If you have not already done so you may want to test out what
performance difference is with server_persistent_connections ON.
* And also with the QUERY acl removed.
* There is a bit of code that can be tweaked to increase the per-cycle
throughput on body data:
src/http.cc line ~1454 change the 1024 minRead to 4096 or 8192 for
less CPU cycles on passing body data.
* Building with --disable-delay-pools since you don't use them will
make squid a little leaner.
* In the same light consider if you need ESI.
Amos
