Subject: Re: New sysctl "net.listen_backlog"?
To: Marc Slemko <firstname.lastname@example.org>
From: Jonathan Stone <jonathan@DSG.Stanford.EDU>
Date: 09/14/1998 14:18:23
>If it has listening sockets, it is a server.
No, it means it's running some server processes. Doesn't make the
machine a "server" by any stretch of the imagination.
Unless you call diskless machines that're running inetd servers? :).
>I understand what you are proposing, and it of course makes sense.
>However, I am amazed that anyone can think it reasonable to have a default
>of 5. I also don't see the point of making it tunable but still having a
>default hard limit as low as 32.
I'm totally confused. The _default_ hard limit, SOMAXCONN, is
currently 128, and has been since at least 1.3.2. Even the old SunOS
4.1. boxes I still have access to are still running quite happily
with an SOMAXCONN hard limit on queuelength of 5. So I'd guess that 32
is a very generous value for many people.
I have to admit I dont understand the idea of 32 as a default; i'd use
the compile-time value of SOMAXCONN as an upper bound on the hard
limit, and use a sysctl to tweak the limit used in uipc_socket.c
between 1 and SOMAXCONN. Net overhead, one variable reference instead
of a manifest constant, plus another sysctl mib number.
If we do that, I dont understand the objection.
If you don't want a smaller limit, don't use the new knob.