Subject: Re: New sysctl "net.listen_backlog"?
To: Todd Vierling <firstname.lastname@example.org>
From: Jonathan Stone <jonathan@DSG.Stanford.EDU>
Date: 09/14/1998 18:28:17
>Let me rehash so there's no confusion about what I've been saying:
>- SOMAXCONN = INT_MAX to request the highest value the kernel allows
> (listen(2) will shrink to fit WITHOUT error)
>- the limit in-kernel, before changing by sysctl, defaults to 128
> (or a compile-time option, if specified)
>- admin can change the in-kernel limit via sysctl to anything up to INT_MAX
>If the admin shoots himself in the foot in the third part above, he shoots
>himself in the foot; we shouldn't try to figure out a "sane" value there.
Sure. I'm just observing we need the second part.
I think we have this beaten to death now, except for sominconn.
I still don't like Perry's suggestion of leaving SOMAXCONN at 128 and
using Matt/DU's sominconn as the way to get round it. What I dont like
is the effect of sominconn on apps that really are single-connection
apps (like ftptd opening a passive-FTP socket) which specify a backlog
of 1. Otherwise, I like sominconn. Comments from the Macadamia stalls?