Re: profiling of a very simple case

Fred CHAUSSUMIER wrote:
>
> Hello,
>
> I would like to know how squid behaves in the very simple case where the
> request is HIT. There are, I believe, a few steps like:
> - parsing a the HTTP request
> - searching in the metadata through the hash table whether the data
> are in the cache or not
> - getting the data from the disk
> - building the HTTP response
> - sending it to the network
> If those steps are almost correct, I wold to know if someone has ever
> done measurements of the time taken by each of these steps especially
> the search in the hash table depending on the size of this one.
> As a matter of fact, I could not find these informations anywhere.
> Thanks,
> Frederique.

Hi Fred,

Alex and Valery Soloviev wrote "On performance on caching proxies",
which is and in-depth perfomance analysis of how squid behaves under
real world load. It identifies and quantify "network and disk
performance degradation during high load periods". You can get and idea
on what time is spent for network i/o, disk i/o, proxy resp. times,
etc... It is not as much detailled as you would like, but it really
deserves a look.

May be you can grab the profiling patch and hack it so it suits your
needs.

Even if it focuses on squid version 1.1.x, I suppose the conclusions are
still true for more recent versions. But you'd better ask this to an
expert.

Paper, patch and docs :
http://www.cs.ndsu.nodak.edu/~rousskov/research/cache/squid/profiling/papers/

Hope your holidays were fun ;), and hope the above helps.

-- 
Michel Blanc <mblanc@erasme.org>
Centre Erasme
Parc d'activit�s innovantes
69930 Saint Cl�ment-les-places
tel +33-(0)4-74-70-68-40 ext 374
fax +33-(0)4-74-70-68-41