Re: compressed TIME-WAIT to be decomissioned

From: Gleb Smirnoff <>
Date: Thu, 13 Jan 2022 16:37:02 UTC
On Wed, Jan 12, 2022 at 11:16:09PM -0800, Chris wrote:
C> > * Who told that 2*MSL (60 seconds) is adequate time to keep TIME-WAIT?
C> >   In 71d2d5adfe1 I added some stats on usage of tcptw and experimented a bit
C> >   with lowering net.inet.tcp.msl. It appeared that lowering it down three
C> >   times doesn't have statistically significant effect on TIME-WAIT use 
C> > stats.
C> >   This means that the already miniscule number of TIME-WAIT connection on a
C> >   modern HTTP server can be lowered 3 times more.  Feel free to lower
C> >   net.inet.tcp.msl and do your own measurements with
C> >   'netstat -sp tcp | grep TIME-WAIT'.  I'd be glad to see your results.
C> I think that should be:
C>      'netstat -sp tcp | grep TIME_WAIT'
C> fe; on the system I'm writing this from:
C> up 15:19, coffee#
C> netstat -sp tcp | grep TIME_WAIT
C> 5 connections in TIME_WAIT state

I'm talking about statistics that I recently committed to CURRENT only:

# netstat -sp tcp | grep TIME-WAIT
        3 times connection in TIME-WAIT responded with ACK
        0 times connection in TIME-WAIT was actively recycled
        0 times connection in TIME-WAIT responded with RST

They show were the TIME-WAITs actually used.

Gleb Smirnoff