At Thu, 6 Mar 2008 12:19:50 -0500 (EST),
Salman Abdul Baset wrote:
> 
> 
> The client discussion has received much discussion. Below, I want to 
> summarize what I think are the requirements for a client which have been 
> previously deliberated but not necesarily put together.
> 
> I will define client as a peer subentity that does not process incoming 
> STORE, FETCH, and FIND messages.
> 
> -Like a peer, a client has an ID. A client obtains its ID in the same 
> manner as a peer.
> 
> -Over its life time, a node may not be permanently designated as a client 
> or a peer. Although, for certain low-end mobile devices this distinction 
> may be permanent.
> 
> -A node may join the overlay as a client. Later, it may be invited by a 
> peer to upgrade itself to the overlay. A client may also decide to upgrade 
> itself to a peer.
> 
> -A node's attempt to join as a peer may be defered by a peer because 
> it has not been up for certain time. The peer can then ask the node to 
> join as a client.

I don't see the point of either the peer refusing to let you join
as a peer or asking you to join as one. The client knows its
own properties better than the peer does and the peer has no
special standing--it just happens to ahve a peer-id 
close that of the client.


> -A client may not be a 'dumb terminal' hanging from a peer. It can provide 
> TURN service. However, it cannot process incoming STORE, FETCH, and FIND 
> messages.
> 
> -A client may be connected to multiple peers for fault tolerance purposes. 
> A client can connect to multiple peers in an overlay (DHT) protocol 
> [in]dependent manner. A client may also maintain a list of other peers 
> without necessarily connecting to all of them.
> 
> -Peers need a mechanism to balance the number of clients connected to a 
> peer.

I think this depends on whether clients attach to peers as a service,
as in RELOAD-03 with explicit routing, or simply as a peer that
doesn't JOIN, as in the non-explicit RELOAD client mode. Unless I've
missed something, this doesn't impose more load on the adjacent
peers (provided that the client creates a robust set of fingers
as it should) than if the client were a peer.


-Ekr
_______________________________________________
P2PSIP mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/p2psip

Reply via email to