Given a chord ring of 100 - 200 - 300 - 400 -
500 - 600 (back to 100) and considering to fight a high churn rate of
400 and 600, than a "new ID" for 500 (e.g. somewhere around 150) would
be good for 500, but wouldn't be of help for the whole topology.
Similarity: You are in a cinema and unfortunately sitting between two
persons, constantly talking to each other. You could choose a new free
seat somewhere off the middle, where you have been sitting up to now.
That would probably help you, but wouldn't make the situation much
better for the whole bunch of visitors :)
Of course the hyperventilation of 400 and 600 might be annoying, but
the ring has to have the ability to always stabilize and cope with
this. A new ID - don't know, where this privileged seat should be come
from :) I'm not aware of having read about such means anywhere.
Regards
stas at khirman.com schrieb:
IMHO Lin Xiao has a quite valid point. If due changes of topology, node 500 permanently loose its ability to interconnect with close ( in DHT space) nodes, it has to be excluded from DHT ring or have to get a new ID.
Otherwise, it will became a permanent dead-end node for lookups for targets in its immediate neighborhood that significantly degrades speed of DHT lookup and ( under some conditions) can event result in DHT space segmentation.
500 can of course keep it's node id, regardless of whether 400 and/or
600 are there or not. Due to stabilization and by means of the
successorcache the predecessor of 400 will notice the absence and bridge
directly to 500 as well as 500 will skip 600 after a short time and go
to 700 (in order to stay in the picture). Using notify the ring will
close again. 400 and 600 may return whenever they want.
Regards
Xiao, Lin (NSN - CN/Beijing) schrieb:
Hi,
Maybe I'm wrong, but I consider peers in RELOAD are structured with a
DHT algorithm, e.g., in Chord, peer with Node-ID 500 is connected with
peer 400 and 600.
However, what will happen, if peer 500 totally lose both of the
connections due to mobility and NAT, but can set up new ones with peer
100 and peer 300? IMHO, in order to keep the topology structured, if
this node still want to be a peer in the overlay it should change its
Node ID to, say 200, and rejoin the overlay to enable itself still
reachable with the DHT algorithm.
If peers in a overlay can not be connected with a strict order, the DHT
algorithm for structured topology will not work anymore.
Br
Xiao, Lin
-----Original Message-----
From: ext jc [mailto:julian at orchidseed.org]
Sent: Thursday, October 22, 2009 12:41 AM
To: Xiao, Lin (NSN - CN/Beijing)
Cc: ext Cullen Jennings; P2PSIP WG
Subject: Re: [P2PSIP] New versions of RELOAD and sip draft
The draft doesn't state that you MUST change you NodeID if you lose
overlay connectivity. In fact under your given condition you would
simply retain it and connect again as a Client when the network is back
up.
Julian
On Oct 21, 2009, at 3:59 AM, Xiao, Lin (NSN - CN/Beijing) wrote:
Hi Cullen,
Thanks for your reply.
I agree that RELOAD have covered most of the functionalities of peers
and clients. However, there is still a little different between
clients and peers while connection is lost besides data replica. What
I mean here is the Node ID.
Because all the peers in the overlay are connected according to a
strict overlay algorithm. The position of a peer in the overlay is
decided by its Node ID. If a peer lost its position (disconnected) in
the overlay because of moving and NAT, but still want to work as a
peer, it must change its Node ID, which means it need do the
enrollment again and rejoin the overlay.
But, because a client can attach with an arbitrary peer. It is not
necessary for a client to change its Node ID when it change its
attaching peer. The most important thing here is that, this behavior
make it possible for a client to keep session continuity while its
attaching peer is switched.
Considering to deploy RELOAD in reality, stable nodes are preferred
to work as peers while mobile nodes can take the advantage of the
client behavior mentioned above. Sessions can be kept with high
mobility, as unique Node ID is maintained by clients. I'm not sure if
RELOAD has considered this distinction of client, which help clients
to be attached with the ovelay as long as it can build at least one
connection to the overlay.
Br
Xiao, Lin
-----Original Message-----
From: ext Cullen Jennings [mailto:fluffy at cisco.com]
Sent: Friday, October 16, 2009 9:05 AM
To: Xiao, Lin (NSN - CN/Beijing)
Cc: P2PSIP WG
Subject: Re: [P2PSIP] New versions of RELOAD and sip draft
Thank you, few comments inline....
On Oct 10, 2009, at 2:20 AM, Xiao, Lin (NSN - CN/Beijing) wrote:
Hi Cullen,
I've had a quick review of these two drafts, and still not quite
clear
with some Client related issues.
I know Client is not the first-class concept in RELOAD, but as it has
been involved in the RELOAD protocol, RELOAD should give full support
to client anyhow, right?
In the terminology chapter of RELOAD Base draft, the connection table
is defined as:
"The set of peers to which a node is directly connected. This
includes nodes with which Attach handshakes have been done but which
have not sent any Updates."
although, the second sentence implies that clients are also involved
in, I still suggest to replace the "peers" in the first sentence with
"nodes" to make it more clear that both peers and clients can be
maintained in the table.
I agree. I changed this - thank you for the careful review to catch
things like this.
In my opinions, connection setup and maintenance between client and
its connected peer (responsible peer or arbitrary peer) is a key
issue for client support in RELOAD. My question is: Does a peer in
RELOAD distinguish if the connection is to a peer or to a client?
Because the reaction of the peer could be different when the
connection is lost.
Reload does distinguish at some level but what do you think a node
needs to do differently when a connection is lost if the connection
when to a peer or client. Clearly how data replication is handled and
some of the finger table stuff is handled differently but I think that
is covered in the draft. If you see something specific missing, let me
know.
As a client could set up a connection only with an arbitrary peer.
Does
this peer distinguish such connection from those it is responsible
to?
I don't se the need to.
Should this peer inform the client's responsible peer about the
Destination List to the client? So that it can still be accessed by
other nodes. Or should the Destination List been stored with some
usage, say SIP Usage?
Yes, that is the approach to store destination lists where they are
needed.
If so, the SIP usage must be extended to allow a Destination list
containing more than one Node IDs.
agree that it needs to allow storing multiple Node IDs but I think it
already does that.
In current SIP
usage draft, the SipRegistration structure only allow one address
being stored in the Destination List, as:
" Destination destination_list<0..2^16-1>;"
The destination_list here is an array of Destinations so it allows
between 0 and 65K entries in the array. I agree the syntax for
specifying arrays might seem a bit weird but I think the structure is
what you are pointing out we need to have.
It should be replaced by: "Destination destination_list
[dest_list_length];"
to enable a longer list stored by arbitrary connected clients.
Anyway, IMHO, RELOAD should make sure to cover all situations a
client could meet or at least clearly distinguish issues of clients,
especially for arbitrary connected ones, which are left to be solved
by other drafts. Thanks.
Best Regards,
Xiao, Lin
-----Original Message-----
From: p2psip-bounces at ietf.org [mailto:p2psip-bounces at ietf.org] On
Behalf Of ext Cullen Jennings
Sent: Wednesday, October 07, 2009 1:53 AM
To: P2PSIP WG
Subject: [P2PSIP] New versions of RELOAD and sip draft
I just submitted
draft-ietf-p2psip-base-04
and
draft-ietf-p2psip-sip-02
These contain technical changes but do not have a lot of editorial
changes. At some point we need to go and reorganize the documents
with editorial changes.
Cullen <in my individual contributor role>
_______________________________________________
P2PSIP mailing list
P2PSIP at ietf.org
https://www.ietf.org/mailman/listinfo/p2psip
_______________________________________________
P2PSIP mailing list
P2PSIP at ietf.org
https://www.ietf.org/mailman/listinfo/p2psip
_______________________________________________
P2PSIP mailing list
P2PSIP at ietf.org
https://www.ietf.org/mailman/listinfo/p2psip
_______________________________________________
P2PSIP mailing list
P2PSIP at ietf.org
https://www.ietf.org/mailman/listinfo/p2psip
_______________________________________________
P2PSIP mailing list
P2PSIP at ietf.org
https://www.ietf.org/mailman/listinfo/p2psip
|