Hi,

Sorry for late response and not to fit an RRG charter.

I believe the host number scalability issue in HPC network
are not different from DCN/DCI.  HPC has a roadmap including
some the number of hosts to be connected, while the
requirements of future DCN/DCI is not clear.

If L2 can be resolve the issues alone, it is not an IETF/IRTF issue.
If we need some support for ARP/DHCP presented as SEATTLE,
EtherProxy, IRTF should do.

HPC network also shares with DCN/DCI in the routing on
large number switches. RBridge and other link-state based
L2 routing can help. However, HPC network topology should be
free from "backbone", since HPC network topology and traffic
have highly symmetrical.

I think many people agrees with the host number scalability issue,
since many research challenges are ongoing and industries may support it.
The later routing issue is only my own idea, and I cannot find any
research paper discussing it other than ATM PNNI. 
So, I don't insist that RRG should pick it.

----
Katsushi Kobayashi

On 2011/05/13, at 14:13, Tony Li wrote:

> 
> On May 12, 2011, at 7:52 PM, Katsushi Kobayashi wrote:
> 
>> I am not sure whether the system will choose regular direct
>> interconnection network such as 3-D torus, or irregular indirect
>> network. DCN/DCI requirements clearly meet the later system.
>> Even in case of direct interconnection, I believe some topology
>> finding mechanism will be strongly required for optimal
>> performance and for operation. 
> 
> 
> Interesting.  As you prepare your topic proposal, please highlight the IP 
> routing and addressing aspects of the research here.
> 
> Thanks,
> Tony
> 

_______________________________________________
rrg mailing list
[email protected]
http://www.irtf.org/mailman/listinfo/rrg

Reply via email to