From mboxrd@z Thu Jan  1 00:00:00 1970
From: ron minnich <rminnich@lanl.gov>
To: 9fans@cse.psu.edu
Subject: Re: [9fans] the backporting continues - /. stories maybe of interest
In-Reply-To: <3ED3516F.90203@proweb.co.uk>
Message-ID: <Pine.LNX.4.44.0305270848360.14285-100000@maxroach.lanl.gov>
MIME-Version: 1.0
Content-Type: TEXT/PLAIN; charset=US-ASCII
Date: Tue, 27 May 2003 08:58:23 -0600
Topicbox-Message-UUID: b982fbee-eacb-11e9-9e20-41e7f4b1d025

On Tue, 27 May 2003, matt wrote:

> Remote Direct Memory Access Over IP

such an old idea. Let's see, how old? Well if you don't flame me for being
involved in such terrible ideas, consider this:

- node exports NFS file system
- exporter and client mmap same file
- client stores to its mmap'ed file, does an msync, memory in
  process on server changes
Voila: remote memory access over IP

OK, enhance it:
- make NFS coherent with a few mods
- put a 'fast bypass' in the ethernet interrupt handler so that, in
  special cases, you bypass the nfs stack and put network data into the
  process memory directly.
QED: MNFS, do a google and you'll find it.

Ok, you don't want the interrupt handler in your way, right? You just want
to dump your data into that procs memory, but still over IP over
something.

OK, google MINI for an interface we built long ago that does this too
-- over IP over ATM, and ATM was the big mistake, but we believed DARPA
:-(

It's still pretty doable. People are acting like this is the Second
Something or something. Yikes.

Plan 9 guys never made the mistake of doing this stuff. It's basically
dumb to worry about RDMA over lines that are 10s of ms in length.

Oh, yes, for real fun, google the Japanese researchers who succeeded in
putting a real NIC into the DIMM slots. Very cool.

You want to make your way in the CS field? Simple. Calculate rough time of
amnesia (hell, 10 years is plenty, probably 10 months is plenty), go to
the dusty archives, dig out something fun, and go for it.

It's worked for many people, and it can work for you.

ron