"Lightfoot.Michael" wrote:
> Hmm, it's still a particularly stupid number on modern networks. It
> makes much sense to set this rather than suffer hundreds or thousands of
> connections sitting there waiting for Godot. :-) BTW, which RFC defines
> this?
Both yes and no. It may be seen as stupid in some networks, but there is
good reasons to why TIME_WAIT exists.
I do not know where this 2 minutes is officially defined in the
standards (the standards say 4 minutes), but it is based on the
TIME_WAIT discussions of STD 7 / RFC 793 "Transmission Control Protocol"
and STD 3 / RFC 1122 "Requirements for Internet Hosts -- Communication
Layers".
In the standard TIME_WAIT is defined to 2*MSL and MSL as 2 minutes. From
what I can tell the general consensus is that 4 minutes is a bit too big
and that half of this MAY be acceptable. Some OS:es use lower values
like 1 minute (i.e. Linux).
> Can't remember offhand but I seem to remember either 3000 or 4000 - and
> I can't quickly find which obscure parameter to ndd is relevant. :-(
Well.. tuning the number of "unbound ports" has a much more effective
impact than tuning the TIME_WAIT timeout.
Regards
Henrik
Received on Wed Nov 27 2002 - 04:01:25 MST
This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 17:11:36 MST