[openib-general] how to choose appropriate ULPs for application

[Caitlin Bestler]

I would also recommend studying DAFS, and in particular
the differences between DAFS and NFS over RDMA.

NFS over RDMA leaves most of the existing file system
logic intact and basically optimizes the exchange of
individual messages over RDMA.

Using SDP would be a somewhat similar strategy,
with slightly less optimizations. The key benefit being
minimal disruption of existing file system layer code.

DAFS would be an example of fully optimizing the
peer-to-peer communication to optimize for RDMA.
If your application lacks an existing intermediate
layer, such as RPC, that type of approach may
be required to get the full benefit of RDMA. Such
an approach would suggest the use of kDAPL.

But I would recommend against use of kDAPL if
you are only replacing the transport of messages
locked to socket semantics. If you are going to
keep the socket semantics then definitely use SDP.


On 7/6/05, 王锡贵 <wxg at ict.ac.cn> wrote:
> Hello！
> 
>     We plan to run lustre on InfiniBand. But now lustre does not support IB.
> I know noting about the IB stack. So I think the IB Upper Layer Protocol(SDP, DAP) could be used to transport data for lustre.
>         But there are too many ULPs over IB. I know little about them. As far as I konw, iSER/SRP is for SCSI command transportation. So, it seems both SDP and uDAP are OK for my plan. I cant tell the difference between them. Which one is better, SDP or DAP? I don't konw how to choose?
> 
>         Any suggestion is appreciated!
> 
> 
> 
> 致
> 礼！
> 
> Sincerely yours
> Wang Xigui 王锡贵
> 2005-07-06
> --------------------
> Computer Architecture Laboratory
> Institute of Computing Technology
> Chinese Academy of Sciences
> Beijing,P.R.China
> Zip code: 100080
> Tel: +86-10-62564394(office)
> Email: wxg at ict.ac.cn
> 
> 
> 
> 
> _______________________________________________
> openib-general mailing list
> openib-general at openib.org
> http://openib.org/mailman/listinfo/openib-general
> 
> To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
> 
>