PHP vs. Python vs. Perl

[Andreas Vierengel]

Steven Grimm wrote:
> I think this is really a test of libmemcache vs. Python vs. Perl. From 
> what you've described, it sounds like PHP's performance was basically 
> not a factor in the results since all the work was being done by the C 
> memcached client library.
> 
> My guess is that if you were to wire libmemcache into Python or Perl 
> (say, using SWIG) you would see a similar speed increase there too. 
> Based on your other results, you'd probably be back to mod_perl being 
> the fastest of the three.
> 
> We actually use the native PHP client with a bunch of local 
> modifications rather than the libmemcache-based one. That's for a few 
> reasons: when we were first setting up memcached, libmemcache was too 
> unstable (which seems to no longer be the case), we want to be able to 
> do rapid test cycles on changes to the client library, and finally, our 
> profiling shows that the memcached client code represents a tiny 
> fraction -- in the 1% range -- of the total CPU time we take to display 
> a page.
> 
> -Steve
> 

that's exactly my experience, too. We use also a self written client but 
in perl, which is about 70% the speed of Cache::Memcached. It doesn't 
show up in our profiling either, so this is was not a major decision for 
us, which language to use or which language is "better" :)
We have one exception and here we use a minimalistic self-written 
c-library to "stream" data to memcache as fast as possible ( about 
20.000 set/s and 2000 get/s with our perl-client from about 30 Webserver )

--Andy