[Beowulf] raw ethernet

Anand Vaidya anandv at singnet.com.sg
Wed Jul 21 07:04:08 PDT 2004


Have you considered M-VIA as a starting point? I did try that earlier, but 
since it is unmaintained ( only selected ancient models of 100M nics work)

Regards,
Anand


On Wednesday 21 July 2004 18:15, 050675 at student.unife.it wrote:
> > Do you mean using raw ethernet packets on a flat bridged network to do
> > point to point IPCs in a cluster application?  No TCP/IP?
>
> Yes, now I'm studying the effects of the raw communication on a
> point-to-point connection. I have two testing nodes disconnected from the
> rest of the cluster and connected one each other with a point-to-point
> connection. The two nodes are P4 pentium with gigaethernet NICs by intel.
> The driver I'm using is e1000-5.2.51 and the kernel version is 2.6.7
> recompiled in order to non-add everything that isn't useful for my
> experiments. Of course I've reliability problems because raw ethernet is
> very useful to increase throughput and to send data as faster as possible,
> but it's not tcp.
>
> > Sounds like a lovely project.
>
> Yes, it's very interesting
>
> > Why don't you report to the list once a
> > month or so.
>
> No problem, I can report my results also once one or two weeks if someone
> is interested and if I have news :-)
>
> > Just remember, while you can likely skip the IP part
> > (which is primarily associated with name resolution and routing) and can
> > maybe skip the port part (dedicated function packets) you'll still have
> > the rest of TCP to recreate -- reliable packet transmission and
> > sequencing and checksumming.  Pretty ambitious, if you're planning to
> > write the whole communications stack yourself (but yes, a great learning
> > project).
>
> Now I'll try to find a stable situation wich lets me have a good
> throughput with very little frame-loss (so interrupt coalescence, NAPI,
> e1000 params tuning and so on)
> I don't know if someone else is making my studies but I would like to
> share ideas (in internet I didn't found many news about raw ethernet, so I
> think that is a quite new mechanism, interesting to develop)
>
>
> Simone Saravalli
> University of physics - Ferrara - Italy
>
>
>
> _______________________________________________
> Beowulf mailing list, Beowulf at beowulf.org
> To change your subscription (digest mode or unsubscribe) visit
> http://www.beowulf.org/mailman/listinfo/beowulf



More information about the Beowulf mailing list