[465] in SIPB_Linux_Development
Resnet causes net card transmit timeouts
daemon@ATHENA.MIT.EDU (Erik Nygren)
Mon Feb 21 15:32:42 1994
To: resnet-help@MIT.EDU
Cc: linux-dev@MIT.EDU
Date: Mon, 21 Feb 94 15:32:19 EST
From: Erik Nygren <nygren@MIT.EDU>
With the 3Com 3C509 under Linux, the following error is generated
during large transmissions:
Feb 21 14:35:57 foundation kernel: eth0: transmit timed out, tx_status 00 status 2000.
A quick poll of 6 users of Linux and 3c509's showed that 5 of them
have this problem. No data gets lost, but a delay is introduced as
the card recovers from the error. One AT-1700 user reports getting an
equivalent error, although this has not been verified. During an ftp
of a few megabytes from the Linux machine to another machine, transmit
timeouts occur a few times.
Thinking it was a driver problem, I talked to Donald Becker (an MIT graduate
and the author of many of the Linux networking drivers). He suggested
that the timeout might be set to low in the kernel. I tried numerous
different values. I even increased it to ridiculously high amounts of time
and the error persisted. Writing back to him with this, he responded:
> The timeout check is there because most ethernet controllers can hang. The
> 3c509 is a recent design by people that have a lot of experience, but the
> people doing the BSD 3c509 driver ran into a similar problem, so I don't
> think it's the driver. It don't happen on all networks, so it must be
> timing related.
>
>The AT-1700 has a similar problem, but it can't always be reset :-<.
From this, it sounds like the problem might be a timing problem
with how the dorm networks are configured. Is there anything
you know of that might be causing this problem?
Thanks,
Erik Nygren
