Re: Help in DSM design

From: Richard B. Johnson (root@chaos.analogic.com)
Date: Fri Mar 03 2000 - 18:13:22 EST


On Sat, 4 Mar 2000, Richard Gooch wrote:

> mauro@alusis.ucb.edu.bo writes:
> > Hi friends!
> >
> > I'm trying to develop a Dristributed Shared Memory sytem(DSM) using
> > Linux. I was wondering myself if using IP for the protocol of the
> > comunicationg nodes were correct?. I thing using IP will add so much
> > bytes to the messages(actually packets) sent by the nodes. Will be
> > better try another lighter protocol?. Could you also point me to
> > other similar projects that are being develop? it will clear my
> > ideas and help me with my problems
>
> Ach! Not another DSM project :-( Don't do it. There's already a DSM
> implementation for Linux, and besides, you're better off with a
> message-passing interface. That way application coders can see how
> costly operations are. Using DSM hides that, resulting in inefficient
> code.
>
> Regards,
>
> Richard....

Here somebody decided that the "ideal" way of doing a project was
to have shared-memory between all the workstations so communications
was "automatic". This was in the 10 mb/s days.

The result was like using a knotted string for memory. Update a single
byte in shared memory and it has to go to all the workstations as
a minimum-size packet. Then all the workstations have to send back
a copy of their new memory image, etc. Necessary locking had the
latency of the network, etc.

This is not something you really should do, even with a light-pipe.
As Richard said, just send messages. It's quick, it's robust, and
no locking issues exist (for the interface).

Cheers,
Dick Johnson

Penguin : Linux version 2.3.41 on an i686 machine (800.63 BogoMips).

-
To unsubscribe from this list: send the line "unsubscribe linux-net" in
the body of a message to majordomo@vger.rutgers.edu



This archive was generated by hypermail 2b29 : Tue Mar 07 2000 - 21:00:27 EST