Chris Wedgwood wrote:
> On Tue, Jan 23, 2001 at 05:50:34PM -0800, Jordan Mendelson wrote:
> Apparently, the performance problem with Win98 <-> 2.4.0 only
> happens with specific ISPs. The MTU and SACK settings didn't
> appear to have any effect.
> Can you easily get packet dumps from both good and bad situations?
I've created 2 sets of tcpdump logs. One using 2.2.16 which represents
the "good" situation. The other from 2.4.0 representing the "bad"
The client in the logs is a Windows 98SE machine dialed up with a local
bay area Earthlink number. The 2.2.16 and 2.4.0 logs were both created
during the same session.
The logs show the Napster sign on process. When you first connect to the
Napster service, the client sends a login string. After the server
authenticates the user, it sends an acknowledgement plus the message of
the day. The message of the day is rather large, but there exists no
throttling of it on the backend. It is sent as a set of send() calls,
one for each line with non-blocking sockets. The client is rather
inefficient in it's read handling; two recv() calls per command are
With 2.2.16 and the Windows client, the MOTD is sent in under a second
and displayed in the client. With 2.4.0, the MOTD takes several times
longer to be fully transmitted to the client. There are multiple
retransmits, apparently corrupted packets, etc.
I did a traceroute from the windows client to the Napster servers to get
the IP of the terminal server the Windows client was dialed into, but
unfortunately it didn't respond to the Windows tracert, so I can't run
nmap on it to check the OS.
I've placed the logs at http://jordy.napster.com/linux/ to save