RE: Whither the RDDP WG

 

> -----Original Message-----
> From: [email protected] [mailto:[email protected]] On 
> Behalf Of Jim Pinkerton
> Sent: Wednesday, October 12, 2005 6:45 PM
> To: [email protected]; [email protected]
> Subject: RE: [rddp] Whither the RDDP WG
> 
> 
> I would suggest that we let the work group remain idle for a 
> year or so, and then re-convene on the next round of 
> innovation. Areas of interest
> include:
> 
> - extending the protocol to support atomics
> - extending the protocol to support concepts of collective 
> operations (aka MPI).
> - extending the protocol after iSCSI/NFS work is done and we 
> better understand the bottlenecks that they encountered so 
> that we can go for a version 2 of this protocol. 
> 
> One of the things that is worthwhile pointing out is the RDMA 
> Read creates huge opportunities for new innovation 
> (generically fetch&op).
> Atomics are obvious. I expect other areas will come up as 
> applications begin to better understand this new paradigm.
> 

And RDMAP/DDP as protocols are very well structured for
extensibility.  The are more then enough unused opcodes
and even more unused queue numbers. The model of sending
an untagged message to a special queue number should
allow for easy protocol extension.