Message-ID: <3A566B96.BCAC86E@napster.com>
Date:   Fri, 05 Jan 2001 16:49:26 -0800
From:   Jordan Mendelson <jordy@napster.com>
Organization: Napster, Inc.
MIME-Version: 1.0
To:     "David S. Miller" <davem@redhat.com>
CC:     netdev@oss.sgi.com
Subject: Re: Mediocre TCP Performance 2.4.0 <-> Win98SE PPP
References: <3A5665CB.30AB2158@napster.com> <200101060015.QAA07335@pizda.ninka.net>
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Sender: owner-netdev@oss.sgi.com
Precedence: bulk
Content-Length: 1491
Lines: 37

"David S. Miller" wrote:
> 
>    Date:   Fri, 05 Jan 2001 16:24:43 -0800
>    From: Jordan Mendelson <jordy@napster.com>
> 
>    I reported this a couple of months back with 2.4.0-test10 and
>    hoping that it might have been fixed I tested 2.4.0 proper with no
>    luck.
> 
> Jordan, if I recall correctly, last time we were analyzing dumps about
> this problem from you Alexey and myself both had concluded that Linux
> was doing things perfectly fine and therefore there is nothing to
> "fix" on the Linux side.
> 
> If I noted a different conclusion at that time, please let me know.

I think the general conclusion was that the Linux side looked like it
was doing the right thing and it was the Windows side or the upstream
ISP that was causing the problem. During this latest test, a different
ISP was used just to make sure (Earthlink) and came up with the same
problem.

Unfortunatly, the number of users who reported having this problem are
too great for me to ignore and I'd imagine anyone running a Web server
would have the exact same problems with Win98 clients dialed up over PPP
w/ compression.

My best guess is that the problem lies with the size of the packets
themselves as the 2.4.0 trace shows the Windows side choaking on the 536
byte packets repeatedly and small writes going through while the 2.2.16
trace shows really only smaller writes. It was mentioned before that
2.2.x packetizes write calls, so this might make sense.

Of course, it's just a guess.


Jordan