Re: [tcpm] New I-D

David Malone <> Tue, 13 February 2007 20:18 UTC

Received: from [] ( by with esmtp (Exim 4.43) id 1HH46S-0005Cp-CC; Tue, 13 Feb 2007 15:18:16 -0500
Received: from [] ( by with esmtp (Exim 4.43) id 1HH46R-0005Cj-JX for; Tue, 13 Feb 2007 15:18:15 -0500
Received: from [2001:770:68:ff::1] ( by with esmtp (Exim 4.43) id 1HH46Q-0003X1-1h for; Tue, 13 Feb 2007 15:18:15 -0500
Received: from ( []) by (8.13.4/8.13.4) with ESMTP id l1DKHl83029125; Tue, 13 Feb 2007 20:17:47 GMT (envelope-from
Message-Id: <>
To: Mahesh Jethanandani <>
Subject: Re: [tcpm] New I-D
In-Reply-To: Your message of "Tue, 13 Feb 2007 10:24:19 PST." <>
From: David Malone <>
Date: Tue, 13 Feb 2007 20:17:47 +0000
X-Spam-Score: -2.8 (--)
X-Scan-Signature: 0bc60ec82efc80c84b8d02f4b0e4de22
X-Mailman-Version: 2.1.5
Precedence: list
List-Id: TCP Maintenance and Minor Extensions Working Group <>
List-Unsubscribe: <>, <>
List-Post: <>
List-Help: <>
List-Subscribe: <>, <>

> A new I-D has been posted on the IETF web site.

> "TCP Maintenance and Minor Extensions", Mahesh Jethanandani, Murali Bashyam, 9

This idea seems quite closely related to TCP keepalives, though it
is addressing a different problem. Do you have any suggestions for
a sockets API to these persist timers?

The draft spends a lot of time showing that the application cannot
deal with the problem. I don't buy all the arguments in this section:
for example, an application can usually control the sockbuf sizes
and then either user non-blocking writes or timeouts to determine
if the application at the other end has gone away or not. I'm not
sure how often an application cares if the TCP at the other end is
still alive!

On the other hand, as pointed out, there are situations where there
is no longer any application (data has been handed off to TCP and
the application has exited/moved on). In these cases it could be
quite hard for an administrator to control the resource usage without
something like the timeouts that you describe. I know FreeBSD has
had to introduce some facilities for dealing with these sort of
resource issues (there is a per-process limit on how many sockbufs
can be allocated by a process and there is a "tcpdrop" utility to
force a TCP connection to be closed).


tcpm mailing list