MC concurrency performance?

[Pierre Sahores]

Alex Shaw a écrit :
> 
> Thanks Andu,
> 
> >Your best bet is to actually build the server and see how it works. The
> >performance is pretty good but not industrial grade.
> 
> I wasn't really expecting to process 1000's of requests a second ..
> .. just needed someone to reaffirm my belief in mc :)
> 
> .. my real concern was losing requests
> 
> regards
> alex
> 
> _______________________________________________
> metacard mailing list
> metacard at lists.runrev.com
> http://lists.runrev.com/mailman/listinfo/metacard

Hi,

I'm currently running four mc-based web applications, working (under
linux) as tcp/ip servers waiting for messages, each one on a different
port. The four apps are running on the same PIII 800 server (in console
mode), behind Apache and a php-script-based sockets listener witch is
used to dispatch the right queries to the right mc-based web apps.

The 4 apps are databases front-ends (flat-files and PostgreSQL based
db). 3 of thoses apps are running since 08/2000 and are used to serve
over the internet an average of 120 connections/second in both
read/write modes. I never had a crash (even about the access concurrency
admin) of thoses configs since they are up (and DMZ firewalled).

The last one (mc+postgresql), currently in test, will have to serve 1000
clients over the internet with an average of about 250
connections/second.

Because i'm not running the last issues of mc, including the "liburl"
features, on my servers, here is the key-note about using mc 2.32 in the
internet zone : mc is realy one of the best tools usables to build great
tcp/ip servers but probably not usable "as this" to build sure and
secure http/ftp clients apps. If you need to build mc-based web/vpn
browsers, you will not get the expected results without sending the
metatalk http/ftp requests trough php, perl, python, etc... calls.

The second key-note is that mc is able to manage very cleanly both
clients and servers tcp/ip apps (without having to run behind an http
deamon alike Apache) if you can build your own sockets protocol (see, in
the runrev archives, what Mark Talluto wrote previously about this).

Regards, Pierre Sahores