UDP testing with G-NIC II

David Skingsley david.skingsley@bt-sys.bt.co.uk
Thu Jan 14 05:50:33 1999


Peter,

The transmit timeout may well be a bug either in the card or the driver.

Reasoning......

When tests are carried out where data is sent also received at least
once every TX_RING_SIZE there are no problems.  However when data is
sent only there does not seem to be an IRQ generated to complete the
cycle, which means after all of the transmit buffers have been filled no
more data is sent.

I did get around this problem but I doubt the solution is any good
although you are welcome to try the driver if you wish.

Incidentally I have had very close to 1GB/s from the cards using UDP and
a permiscuos socket.  The was verified using a gigabit router (Always
assuming the counters are honest :-)).

David.

Usual disclaimer about these opinions being my own and not necessarily
that of my company.

>-----Original Message-----
>From:	pmonta@halibut.imedia.com [SMTP:pmonta@halibut.imedia.com]
>Sent:	Wednesday, January 13, 1999 11:23 PM
>To:	yellowfin@cesdis1.gsfc.nasa.gov
>Subject:	UDP testing with G-NIC II
>
>Thanks to all who sent performance information---I've been
>able to reproduce the TCP number of about 368 Mbit/sec on a
>PII/350 (100 MHz FSB), Linux 2.2.0pre6, TCP window=65000
>bytes.  Most cool.  Perhaps I'll try window scaling, though
>it seems like the latency is too low for this to matter;
>probably the CPU is just maxed out.
>
>However, has anyone been able to run tests for UDP, for
>example ttcp in UDP mode?  I get "transmit timeout" errors
>from the driver, but perhaps I'm missing something basic.
>Even a small test like "ttcp -t -u -s -n 64 -l 1400" just
>hangs after a short time, then subsequent traffic (like a
>ping) gives the transmit-timeout errors.  As a pure shot
>in the dark, it could be that TCP's window has prevented
>corner cases in the driver like Tx ring full.  Or perhaps
>I need to increase some kernel UDP or IP resource.
>
>Cheers,
>Peter Monta   pmonta@imedia.com
>Imedia Corp.
> | To unsubscribe, send mail to Majordomo@cesdis.gsfc.nasa.gov, and
>within the
> |  body of the mail, include only the text:
> |   unsubscribe this-list-name youraddress@wherever.org
> | You will be unsubscribed as speedily as possible.
>
 | To unsubscribe, send mail to Majordomo@cesdis.gsfc.nasa.gov, and within the
 |  body of the mail, include only the text:
 |   unsubscribe this-list-name youraddress@wherever.org
 | You will be unsubscribed as speedily as possible.