reset TCP connections to *.microsoft.com

jdm at joshua.enteract.com jdm at joshua.enteract.com
Tue Jun 9 20:14:21 UTC 1998


Mr. Dana Hudes <dhudes at graphnet.com> writes:

> Anyone else experiencing trouble connecting various microsoft www
> and ftp sites? I put a Sniffer on the line and found that I have a
> part of a web page come up and then the connection is reset.  This
> is using Netscape Communicator4 on both NT and SPARC Solaris. If one
> uses IE3 it works better.  I find this highly suspicious. Also
> straight ftp from a unix system is very slow if you get in at all. I
> think we can all agree that the tcp/ip stack on sparc solaris 2.5.1
> is not likely broken (patched to current level, January '98, just in
> case) where NT might be suspect even 4.0 w/service pack 3.

This curious behavior was discussed a few months ago on the mailing
list for the TCP implementors mailing list.

The Windows TCP/IP stack appears to use RST to mean "busy, try again".
Unfortunately, no one else does, nor is this behavior needed.

An NT machine (presumably what is running on Microsoft's www and ftp
sites) will issue an RST on an incoming connection when the socket
queue is full.  An NT machine acting as a client retries a SYN
responded to by an RST 3 times, separated by a half-second, and only then
gives up.

The proper behavior would be for NT to ignore SYNs when its backlog is
full, and rely upon the client machine to retry the SYN.  No other
TCP/IP stack but Microsoft's is going to retry a connection which was
terminated with an RST, so you will likely end up with broken images
on a web page.  It's possible that IE3 handles this situation better
than Netscape because, knowing this quirky behavior, it might attempt
to compensate for it by retrying a connection refused a few times.

--
Jennifer Dawn Myers                                     <jdm at enteract.com>
TCFKASNI                                     http://www.enteract.com/~jdm/



More information about the NANOG mailing list