Skip site navigation (1)Skip section navigation (2)
Date:      Sun, 2 Jul 1995 09:26:13 +0300
From:      Petri Helenius <pete@sms.fi>
To:        jleppek@suw2k.ess.harris.com (James Leppek)
Cc:        freebsd-questions@freefall.cdrom.com
Subject:   Re: tcpblast
Message-ID:  <199507020626.JAA19099@silver.sms.fi>
In-Reply-To: <9507020111.AA00633@borg.ess.harris.com>
References:  <9507020111.AA00633@borg.ess.harris.com>

next in thread | previous in thread | raw e-mail | index | archive | help
James Leppek writes:
 > Pete,
 > 
 > You may have something there.
 > if I leave the mtu at 16384 I get a thruput of 2.1 Mbytes/s
 > but it drops to .1 after a few test. If I change the mtu
 > to 1500 a get a thruput of 1.3Mbytes/s but no stalls.
 >
Sounds more like it. The mecanics of tcp vary from implementation to another
but take a look at ftp://ftp.scs.leeds.ac.uk/atm/doc/tcp-deadlock.ps
for a review what might be happening. The document is not the whole truth,
I've seen the deadlock to appear even more easily. 

 > Would it be better to change the default mtu for lo0 to a smaller
 > value as a default??  If I use an mtu of 6000 the thruput is at 2Mbytes/s
 > with no stalls.
 >
Safe value half of the default window. I think window is 16384 in FreeBSD so
something around 8000 should work best. Some applications do lower or raise
the window to get special loading/throughput requirements filled. Raising
the default window is very bad idea. (not that I'm suggesting that you were
doing it)

 > Hmmm is this thread starting to belong in hackers :-)
 >
I that case, keep me in, unfortunately the general volume of it too much for
me to just watch the networking-related issues...

 > Jim
 > 

Pete



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199507020626.JAA19099>