RE: FW: AD review of draft-ietf-rserpool-comp-07

> ...
>
> As for the other portions, I'm not familiar with them so I will have to 
> read the drafts.  I think that some of the comments that Qiaobing made 
> in the past regarding loosely-coupled approaches of trying to integrate 
> many protocols versus a tightly-coupled protocol (RSERPOOL) will hold 
> here as well.  All the more so if we stress that RSERPOOL is intended 
> for system with some degree of real-time performance constraints.
>
> Aron

What do you mean by systems with "real-time" performance constraints?

If it means to quickly obtain the best server from a pool maybe the
mandatory usage of SCTP is not an optimal solution.

Let's imagine an user that wants to access a web-service served by a
pool of servers. That means that first it has to establish a SCTP
connection to the ENRP server (4-way handshake), get server info, and
finally open a TCP connection (3-way handshake) to the selected web
server for the HTTP transaction. It makes more or less 4,5 RTTs.

It seems that querying ENRP servers via UDP will lead to better
server-selection latency for "real-time" services (2'5 RTTS for TCP
connections).

Regards,
--Manuel

-- 
Manuel Uruen~a - Universidad Carlos III de Madrid
GPG FP: 68A1 164B EE28 52C9 87CB  EBF9 616E 52B5 451A B6B2