Re: [Ecrit] Comments/Questions on draft-ietf-ecrit-lost-02

[Andrew Newton <andy at hxr.us>]

On Nov 21, 2006, at 9:17 AM, Reese, Theresa E wrote:
[..snip..]
> Can you tell me if there been any discussion regarding the  
> inclusion of
> something like a Message ID in the LoST protocol, or whether the
> assumption has been that this correlation would be based on the TLS
> connection over which the specific query/response was exchanged?  Has
> there been any discussion regarding the number of TLS connections that
> might exist between an ESRP and a LoST server to handle potential
> traffic volumes?  (I know this goes beyond the specification of the  
> LoST
> protocol itself, but it does relate to the use of LoST for routing
> emergency calls.)

Since the current definition of LoST uses HTTP and HTTPS, both of  
which are TCP based, the way that a request is matched up with a  
response is via the session semantics of HTTP over TCP.  We have not  
discussed a message ID, as that is really a trait of the transfer  
protocol and its bound transport protocol.  With HTTP, you simply  
rely on the underlying TCP session semantics.

This means that an ESRP would indeed need to keep a connection pool  
of long-lived HTTPS/HTTP connections.  However, because of the  
caching involved and the likelihood that an ESRP would probably be  
coupled with a LoST caching resolver, it is not clear to whom the  
persistent connections will need to be made and how many will be  
needed.  We can theorize that an ESRP/LoST resolver pair will need a  
certain number of connections to the forrest guide of its  
jurisdiction and perhaps a certain set of LoST authoritative servers  
within its coverage area, but I think deployment experience will be  
the only way we really know in the end.

Within the author team and the working group, we have discussed  
bindings to other protocols... specifically SOAP and UDP.  I don't  
believe SOAP offers an answer here (and it may cause problems,  
actually), but UDP would allow many simultaneous sessions if the UDP  
binding had a message ID (like DNS or IRIS-LWZ).  The obvious problem  
with UDP is that the payloads cannot be large and there is no  
practical channel security (I believe DTLS would get you right back  
to the same point as using HTTPS).

We, the author team, decided that we could spend many thousands of  
hours trying to figure this out, and in the end we may not get it  
right due to lack of true deployment experience.  Will we need UPD?   
Will we need something like BEEP?  We truly don't know, but we are  
very confident that the XML as designed is easily adaptable to other  
transfer/transport schemes.

-andy

_______________________________________________
Ecrit mailing list
Ecrit at ietf.org
https://www1.ietf.org/mailman/listinfo/ecrit