[medusa] Re: any performance info?

[Sam Rushing]

Stuart Zakon writes:
> *So I don't have to dig down too deep myself, how do thttpd and
> Zeus achieve greater performance if they use an asynch scheme? Is
> it because they are closer to the kernel?

The async approach allows them to mix I/O with computation. Serving
static pages is one extreme... there's no computation at all, just
network I/O. And it all takes place in 'parallel': making 30
non-blocking connect() calls is the equivalent of saying "connect
these 30 sockets to these 30 hosts, and let me know when they get
through".

> *Has anybody done anything with asynch for reverse DNS lookups?

I think Google uses async DNS lookups in their crawler. Newman (the
eGroups mailing-list exploder) does as well.

> *Is your overview "Programming in Python with Medusa and the Async
> Sockets Library", which I found very useful, directly accessible on a
> web site without having to open the distribution? (Would be helpful for
> linking to from other pages.)

This will work:

http://www.nightmare.com:8080/nm/apps/medusa/docs/programming.html

I think a few other people are linking to it in that location as well.

> *Does your mailing list exploder use the poll() approach?

Haven't bothered, yet.

-Sam