Re: [squid-users] Is Squid slow?

Joe Cooper wrote:
>
> Doh! You're right Vivek. I stand corrected. At the second cacheoff,
> we weren't pushing quite as hard and /did/ show response times among the
> best at the event...only beaten by the iMimic entry and Duane's Squid entry.

Again, it's a matter of looking more closely at why things
performed the way they did. From the second cacheoff, we
can see the hit times at

http://polygraph.ircache.net/Results/cacheoff-2/polyrep/cmpAll/hit.rptm.mean.by-meas.png

and the miss times at

http://polygraph.ircache.net/Results/cacheoff-2/polyrep/cmpAll/miss.rptm.mean.by-meas.png

Lower bars are better in both graphs. In each case, we see that the
Squid-based entries are roughly in the middle of the pack. However,
they score quite well in the overall response time, shown in

http://polygraph.ircache.net/Results/cacheoff-2/polyrep/cmpAll/rep.rptm.mean.by-meas.png

The reason for this is simply because some of the other boxes were
either poorly configured or intentionally driven beyond the loads
they could comfortably handle with good hit rates. For example,
the Dell 130B box scores among the worst in overall time, but beats
the Squid entries in both hit times and miss times. It just suffered
from a really bad hit rate, causing overall time to be worse.

> It's merely a matter of how hard you're pushing the box and the hit
> ratio you're showing...If you push harder (for more throughput) response
> times go up.

But again, there's the caveat that this holds true only if the hit
ratio is unaffected. The Dell box is being driven hard enough to
cause hit rate to suffer, but the individual hit/miss times look fine.
In real life, it wouldn't be wise to run the box this way, simply
because it's not getting all of the possible hit rate benefit.

-Vivek
Received on Wed Dec 05 2001 - 12:02:38 MST