maximum throughput per hour of your squid?

From: Joe Ramey <ramey@dont-contact.us>
Date: Mon, 7 Apr 1997 20:23:49 -0500

We're trying to come up with some metrics that will help us decide
when to add additional proxies to our network. I've seen people
posting their stats of transfers per day, but we find that transfers
per hour make more sense to us, since they seem to more accurately
indicate the most throughput that a given system can do.

In looking at our stats for the past several months we've found that
our Sun SS10 can do around 90000 xfers/hour, our SS20 can do around
95000, and our Ultra 1 can do around 120000 per hour. Do these
numbers seem reasonable? We aren't doing caching on any of them right
now, as we found that our transfer rates went down when we enabled
caching. We tried spreading the cache among 4 disks and still our
xfer rates were lower when the cache was enabled. I want to study
this some more to make sure we hadn't done something dreadfully wrong.
We do have a very fast connection to the Internet, though, so maybe we
get pages in fast enough through the network that caching doesn't help
us all that much? I'll admit that we gave squid very little in-memory
cache to play with since our squid proxy systems also see duty as
e-mail gateways (lots of sendmail processes using lots of memory at
times).

So, first of all, do these numbers seem reasonable as approx. maximum
throughputs? And secondly, how do you decide when your squid proxies
are overloaded and it's time to add another one?

Joe
Received on Mon Apr 07 1997 - 18:31:47 MDT

This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:34:57 MST