Subject: Re: New sysctl "net.listen_backlog"?
To: None <>
From: Michael C. Richardson <>
List: tech-net
Date: 09/14/1998 22:17:07
>>>>> "Jonathan" == Jonathan Stone <jonathan@DSG.Stanford.EDU> writes:
    Jonathan> But to achieve the underlined part with the extant sockets
    Jonathan> codebase, you need to make SOMAXCONN be UINT_MAX.

    Jonathan> So then we _need_ yet another option, to set the default upper
    Jonathan> limit actually enforced by listen() to something sane.  (where
    Jonathan> "sane" is possibly arch-dependent).  Like the knobs Matt
    Jonathan> suggested.

    Jonathan> Is that the wavelength you're on?

  From what I've observed, back here in the bleachers, this makes the
most sense.
  SOMAXCONN is UINT_MAX for applications that couldn't be bothered to
pick something else, and we have two limits -- one for a system hard limit
beyond which you can't tune above, and the other for a default for listen()
which defaults to something reasonable for small-middle sized systems.
  I suspect that the former doesn't need to be a sysctl. The later does.

   :!mcr!:            |  Network and security consulting/contract programming
   Michael Richardson |         Firewalls, TCP/IP and Unix administration
 Personal: PGP key available.
	ON HUMILITY: To err is human, to moo bovine.