Re: Cacheflow vs. Squid

Yeah, for that kind of money it should at least have some blinking
lights.

Got it up and working. After just 5 minutes the reported hit rate
was around %37. After around 30 minutes it reported around %47 with
only around 150 megs in the cache (started clean). This seems to me
to be a bit optimistic. At peak load we may have 400 customers dialed
in and I don't understand how you can get that kind of hit rate on
150 megs of objects just 30 minutes old. I have read their web pages
on their superduper proprietary statistical prediction algorithms,
but I just don't see how it can get enough trend information in
30 minutes from random user access patterns.

I was/am of the opinion that if you start squid with and empty cache
and start cacheflow with an empty cache that the playing field is
going to be level for a least a while. I did find a blurb on their
page about squid always fetching an object from the server on the
second cache hit to insure that it is fresh, and that their nifty
algorithm determined if it needed to be refreshed based on usage
patterns. Just playing around with a browser when squid is running,
I see some human noticeable delays when you hit a new page, and
it comes up quickly many times thereafter (even after clearing the
cache on my browser). I've not been able in their management interface
to collect information on the actual URLs retrieved (I wanted to
see if such a hit rate was even possible based on what was retrieved
in the first five minutes).

I thought it may be fetching several URL's off a site when you hit
a page, but I check the access logs on a remote server, and there
was only an access logged for what I had requested. The growth of
the cache (starting from empty) seems to parallel squid closely.

It performs well thus far. The downside is that with all the
proprietary lock and key stuff, recovering from failure is going
to be messy.

I did find it humorous that the probe for the Intel NIC spits out
a message identical to FreeBSD when it does the same. The tech at
cacheflow was kind of suprised, said that such messages were supposed
to be disabled.

Something we keep hearing from this and other vendors, is that the
"guy" that wrote squid now works for them.

-

B. Richardson rabtter@orion.aye.net

On Wed, 29 Jul 1998, Robert Hof wrote:

> Ah, but you pay for the pretty case ;)
>
> At 08:37 PM 7/28/98 -0400, you wrote:
> >
> >Test driving one at the moment, having got it working right yet. Funny
> >thing, connected a monitor and keyboard to it and saw it probe
> >for fxp0 and fxp1 during bootup and the messages looked *incredibly*
> >familiar :-). Had an adaptec SCSI controller in it (and of course intel
> >NICS :-) ).
> >
>
> Rob
> ----
> Robert Hof,
> Head-Honcho-Type If at first you don't succeed, then
> NETinc - The Internet People skydiving definitely isn't for you.
>
Received on Wed Jul 29 1998 - 11:30:16 MDT