Subject: Re: several messages
To: Perry E. Metzger <perry@piermont.com>
From: Todd Vierling <tv@pobox.com>
List: tech-net
Date: 09/14/1998 20:33:01
On Mon, 14 Sep 1998, Perry E. Metzger wrote:

: btw, I think that "net.inet.tcp.maxconn" or the equivalent is probably
: a better name.

It's a wrong name.  The listen backlog is independent of protocol - it's
only SOCK_STREAM specific.

: > Now, what about an "options LISTEN_BACKLOG" to set a compile-time default,
: > in the same way that DDB_ONPANIC sets the switch at compile time?  (If
: > nonexistent, the option would of course default to 128.)
: 
: I don't like calling it LISTEN_BACKLOG, given that the thing it
: touches is maxconn. The option is a decent idea.

Well, that's what (AFAIR) Solaris calls it, right?

: > And, what should the hard limit be--that even sysctl() can not go above?
: 
: UINT_MAX.
: 
: > (This will become the value of SOMAXCONN such that programs using the
: > macro will gain the higher limits without recompiling.)
               ^^^^^^^^^^^^^^^^^^^^^^^===================
: Leave SOMAXCONN at 128, I say.

Then what use is SOMAXCONN as it is defined in <sys/socket.h>?

-- 
-- Todd Vierling (Personal tv@pobox.com; Bus. todd_vierling@xn.xerox.com)