Mark Elsen wrote:
My College has a T-1 line that is maxed out most of the time. They also
have some wireless lines connected to 1MB DSL lines and a new Cable
Line that is setup at 4.2MB. I have a Fedora Core 4 machine setup with the
Advnced Linux routing to connect the Cable and Wireless to the machine as
default routes with equal weighting. My test lab is set to use this machine as
the proxy, and I can see the load accessing both the wlan0 and the eth, so
that is working.
...
- One, of the problems, I personally feel, is that the Redhat RPM builds
come with all SQUID options enabled, during configure.
This leads to a heavier exe and 'carried code' which isn't used.
I always recomment to fetch the latest STABLE release and make a
build (configure) with those configure options which you need and none other.
This can also help in performance terms.
M.
Be aware that epoll is not part of the STABLE squid sources. It's a
devel package that is only available by patching the sources, and
running a bootstrap script. Depending on the load of the cache, it
might not even help much. You might look into changing the cache_dir to
use aufs, but if cache hits are performing well, this also might not
make much of a difference. Given that cache hits are working well, but
cache misses are slow, I would look into the DNS server that Squid is
using, as well as network congestion on the path to internet.
Chris
