Re: RDMA over TCP (Was Re: VI (Was: Avoiding deadlock in iSCSI))

[julian_satran@il.ibm.com]

Steph,

I made the same points as you a long time ago on this list (before
Adelaide) when Costa came out with an RDMA option for TCP.

I think that at the time I was vocal in support of such an approach.

Watching what happened to it and considering that we have to have
widespread
support to get to use it for storage I think it will be counterproductive
to have iSCSI
wait for it.

That is what "Julo was not buying"



Stephen Bailey <steph@cs.uchicago.edu> on 29/09/2000 00:36:40

Please respond to Stephen Bailey <steph@cs.uchicago.edu>

To:   ips@ece.cmu.edu
cc:    (bcc: Julian Satran/Haifa/IBM)
Subject:  Re: RDMA over TCP (Was Re: VI (Was: Avoiding deadlock in iSCSI))




> I believe the point in factoring out the RDMA mechanism from iSCSI is to
> make the same hardware assists available to other application
> protocols.

I agree.

The motivation is primarily economic.  If you implement a general
solution to 1) data steering, 2) congestion avoidance, 3) security,
you own the edge of all networking.  If you implement these for
storage, you only own storage.

Given that it seems such an easy step to make the hardware slightly
more general and own the world, I'm stumped on why not.  OK, I admit I
know the argument why not, because people think the more specific
solution more tractable.  I disagree.  The other reason why not is
some people really aren't sold on the necessity for hardware
implementation at all.

BTW, I made this case at CERN recently:

http://hsi.web.cern.ch/HSI/HNF-Europe/Workshop%202000/Presentations/Com_Stor_Net.PDF


Julo wasn't buying though.

Steph