scalable network architectures (was RE: [Mono-list] Mono fitness for ircd project)

Matt Liotta mliotta@r337.com
Mon, 3 Jun 2002 21:14:07 -0700


We were talking about scalable in the context of current number of
sockets. Real-time signals allow for significantly more current sockets
than any other architecture.

-Matt

> -----Original Message-----
> From: Michael Poole [mailto:poole@troilus.org]
> Sent: Monday, June 03, 2002 9:12 PM
> To: Matt Liotta
> Cc: mono-list@ximian.com
> Subject: Re: scalable network architectures (was RE: [Mono-list] Mono
> fitness for ircd project)
> 
> "Matt Liotta" <mliotta@r337.com> writes:
> 
> > With proper kernel support real-time signals should provide the most
> > scalable network architecture.
> 
> Scalable by what metrics?  For maximum throughput, you do not want a
> context switch every time there is new activity on a socket -- but
> that is what signals like that give you.  (They will reduce average
> latency.. until you saturate the CPU.)
> 
> Several OSs' /dev/poll and FreeBSD's kqueue() API will return sets of
> N (>=1) active sockets at a time, which helps a lot if you have lots
> of small requests going.  As in IRC.
> 
> -- Michael