I agree that NAT traversal is complex but that is for a peer to peer network 
where both the peers are behind NATs and they need to figure out their 
reachability. In case of ALTO we don't have that kind of complexity. The ALTO 
client which expects the server to connect to it needs to ensure that it is 
reachable on that address/port from an external/internal network. This is same 
as the way an ALTO server behind a carrier firewall is configured to be 
reachable to clients outside.

I don't envisage ALTO clients like home devices using the subscription 
mechanism because, for a home device client it is easier to query the ALTO 
server for a changed map information rather than going for a subscription. 

IMO, this subscription mechanism would be useful for infrastructure cases like 
gateway devices, CDN, inter-DC etc where the problem of reachability is not so 
complex as in the case of peer to peer.

Please let me know if there is something I am missing here.

-Ravi
________________________________________
From: Enrico Marocco [[email protected]]
Sent: Tuesday, July 24, 2012 4:10 PM
To: Ravi nandiraju
Cc: [email protected]
Subject: Re: [alto] WebSocket-based notifications

On 7/24/12 11:57 AM, Ravi nandiraju wrote:
> Yes. Websockets will be useful for cases where there are frequent
> notifications (like server load etc). This is only one use case for
> subscription/notification. IMO, we should define a generic framework
> for subscription/notification which includes support for things like
> multiple event subscription and does'nt have the overhead of
> persistent connections when the clients are not interested to keep
> them.

This is actually something the authors would like to get feedback about,
i.e. whether people prefer to have one connection per resource, or some
sort of subscription protocol running for multiple resources on a single
connection. The draft at this point suggests the former approach,
basically for simplicity.

> Regarding the server creating a connection to the client, the client
> explicitly advertises its contact address for the server to connect
> to. The reachability problem should be solved by the client.

I cannot think of any way for the client to solve the reachability
problem that is less demanding in terms of resources than a
client-initiated connection. Can you suggest one that works with today's
firewalls, access gateways and carrier NATs?

--
Ciao,
Enrico
_______________________________________________
alto mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/alto

Reply via email to