Re: squidfs type thing & benchmarking

On Tue, 20 Apr 1999, Eric Stern wrote:

> Is there someone who can take some prototype code and give it a good
> benchmarking? I thinking a test identical (or really close) to the Bakeoff
> test, since we can then compare the results to those results.

Polyteam can. We've benchmarked several commercial caches and numerous
Squid configurations with Polygraph. And we know the bake-off workload
pretty well, since we've designed it. We are in the process of building
PolyLab, a benchmarking lab for the Web caching community. I hope the Lab
will be operational in May.
 
> I've got my own version of squidfs (except I call it COSS...Cache Object
> Storage System) working, but I'd like to get some benchmarks in before I
> go further with it. Unfortunetly I don't currently have the resources to
> give it a real good test myself.

It all boils down to three things:
        1) the effort required to incorporate COSS into Squid
        2) the equipment you want to test Squid on and with
        3) the expertise to run a benchmark and interpret the results

I assume (1) is not a problem. Polyteam has (3). As for (2), I am sure
PacketStorm is interested in making Squid perform at its best. Thus, it
certainly makes sense for PacketStorm to become a "member" of our lab.
Given that, we can probably start testing COSS and, perhaps, other versions
of SquidFS in May. If you are interested, please drop me a line.

If PacketStorm does not want to contribute, we still can do the tests. The
only difference is timing and the scale of experiments.

A nice and fair comparison of various approaches would be very useful, IMO.

Alex.
Received on Tue Jul 29 2003 - 13:15:57 MDT