Subject: Re: Web load causes reboot
To: Jason Thorpe <thorpej@nas.nasa.gov>
From: Michael Graff <explorer@flame.org>
List: current-users
Date: 08/04/1999 10:03:59
Jason Thorpe <thorpej@nas.nasa.gov> writes:
> On 03 Aug 1999 17:17:17 -0700
> Michael Graff <explorer@flame.org> wrote:
>
> > It's unfortunate that network sockets seem to reserve so much space.
> > How do other 256M machines running FreeBSD (for example) cope with
> > 10,000 or more TCP streams open at once? If we did that, we'd end up
> > allocating about 312M just for socket buffers, I think...
>
> Um...
>
> Socket buffers aren't "allocated". When you "allocate" a socket buffer,
> you're placing a limit on it... the actual space is lazy allocated... the
> "allocated" space is actually mbufs allocated in a network driver and
> handed up the stack.
Ok, let me rephrase. "a few hundred sockets sending/receiving data as
fast as possible" will crash and/or hang any out of the box NetBSD
machine.
--Michael