The fix in all versions of Linux for the last year or so is an easy one:
echo 1 > /proc/sys/net/ipv4/tcp_rfc1337
This sets the Linux TCP stack to obey RFC 1337 rather than what the
Linux kernel folks believe to be the 'correct' implementation with
regard to assassination (which msl_test uses, and causes a normal Linux
TCP stack to return after just 2 seconds). Everyone on both sides
believes theirs is the correct implementation. So it's not really a
bug...just a different interpretation of the standards. And a patch is
not needed to work around that fact.
I have found a somewhat confusing fact about the 2.4 kernel...it will
return anything from ~55 seconds to ~63 seconds when running msl_test
against it. It may be related to the AMD processor we use, as I've seen
other timing related issues, but the 2.2 kernel is not subject to this
fluctuation, so I don't know where it comes from.
More unrelated postings, I know...but I figured if anyone stumbled on
this thread while trying to get the right values out of Linux/msl_test
they'd appreciate not having to trace back all of the various threads on
the subject.
Alex Rousskov wrote:
>
> Steven,
>
> As far as I can tell, the posting investigates a [minor] Linux
> TCP stack bug that leads to 2-second TIME_WAIT values under certain
> conditions. The thread has a patch that fixes the bug. So it looks like
> msl_test was correct in reporting 2 second TIME_WAIT for the Linux
> person who opened the fa.linux.kernel thread.
>
> I believe that the strange behavior we have seen during the
> cache-off is not related to the above thread: The caches in question
> were running FreeBSD rather than Linux and the 2-second TIME_WAIT value
> was due to ARP(?) problems on the cache. Same reported value, different
> cause. As you know, we've got the correct measurement of 60 seconds when
> using Polygraph aliases rather than the monitoring station IP address.
>
> The above URL may be of interest to those who have problems with
> msl_test in Linux environment.
>
> Thank you,
>
> Alex.
--
Joe Cooper <joe@swelltech.com>
Affordable Web Caching Proxy Appliances
http://www.swelltech.com
This archive was generated by hypermail 2b29 : Tue Jul 10 2001 - 12:00:15 MDT