I think both randomness of node ID and proximity-aware DHT are options. Randomness of node IDs is required in some overlays for security and load balancing reasons. Some overlays may not require randomness of node IDs because security and load balancing issues are already handled by other means. Sometimes proximity-aware DHT is not necessary. For example, all nodes of a overlay locate in the same LAN.
On Wed, Dec 3, 2008 at 7:30 AM, Bruce Davie <[EMAIL PROTECTED]> wrote: > Well, I'm one of those " Alto guys ". Alto will answer queries about > proximity, but the question of randomness of node IDs is well outside Alto's > scope. This working group will have to decide whether, as I suspect, > randomness of node IDs is an important property to preserve. > > Bruce > On Dec 2, 2008, at 11:13 AM, songhaibin 64081 wrote: > >> Bruce, >> >> I agree with you that pastry has proximity property, and can achieve >> efficent routing. But for leafset, peer must choose leafset nodes that have >> closest IDs with it. It doesn't have a list of candidate peers to choose >> from as the routing table entries do. SO, by randomly assigned IDs, it will >> make the resource records mainternance between leafset nodes(when join or >> leave, or replicate happens often) not efficient because they waste the >> network hops. >> >> If node IDs have network proximity property, it will make the routing >> message which gets closer and closer to the destination ID(e.g. Chord or >> Pastry), also gets closer and closer to the destination node in the network >> topology. It is a nature effect. >> >> We could discuss with the ALTO guys and see what in their mind. >> >> I think we should allow different options for node id assignment according >> to different scenarios. >> >> my two cents >> Song Haibin >> >> >> ----- Original Message ----- >> From: Bruce Davie <[EMAIL PROTECTED]> >> Date: Tuesday, December 2, 2008 11:39 pm >> Subject: Re: [P2PSIP] P2PSIP ID and physical location >> To: songhaibin 64081 <[EMAIL PROTECTED]> >> Cc: P2PSIP WG <[email protected]> >> >>> Song, >>> it is certainly true that proximity is often used in DHTs, but >>> not, as >>> far as I know, for the assignment of node IDs. It is used, for >>> example, to select among many possible candidate nodes for >>> inclusion >>> in the routing table. But, as others have already said, the >>> randomness >>> of note IDs is important for the security and robustness of a DHT. >>> You >>> can have efficient routing without giving up the randomness of >>> note ID >>> assignment. See, for example, Pastry. >>> http://www.freepastry.org/pubs.htm >>> Bruce Davie >>> On Dec 2, 2008, at 10:11 AM, songhaibin 64081 wrote: >>> >>>> Hi Ekr, >>>> >>>>> >>>>> It's not clear to me why it's desirable for nodes which are >>>>> geographically close to be close in the overlay topology.y >>>>> >>>>> -Ekr >>>> >>>> P2P networks are teemed with dynamic nodes. When a peer leaves >>> >>> the >>>> >>>> overlay, it will transfer the stored resource records to a node >>>> which has a close node ID. From the perspective of an ISP's >>> >>> network, >>>> >>>> making P2PSIP node IDs with proximity property will reduce the >>>> network hops when data transfer happens. >>>> >>>> Besides that, I think overlay maintenance messages will also >>> >>> take >>>> >>>> less network hops if most neighbors have close node IDs. If node >>> >>> ID >>>> >>>> has proximity property, it will help for the peer selection when >>> >>>> there are a list of peers providing the same resource. >>>> >>>> I think we can find some papers describing the proximity for >>> >>> DHTs. I >>>> >>>> agree with Bruce that there are many considerations when >>> >>> assigning >>>> >>>> node id. >>>> >>>> Best Regards, >>>> Song Haibin >>>> >>>> >>>>> >>>>> At Tue, 02 Dec 2008 14:26:56 +0100, >>>>> Xianghan Zheng wrote: >>>>>> >>>>>> [1 <text/plain; ISO-8859-1 (7bit)>] >>>>>> Hello, >>>>>> Is someone considering the mapping between the P2PSIP ID and >>>>> >>>>> physical >>>>>> >>>>>> location. I think it is necessary to think about it although it >>>>> >>>>> is not >>>>>> >>>>>> trivial and might cause some security problem. >>>>>> >>>>>> In the draft >>>>>> "http://www.p2psip.org/drafts/draft-licanhuang-p2psip- >>>>> >>>>> subsetresourcelocation-00.txt", >>>>>> >>>>>> However, the peer ID is formed as domain name. Is that conflict >>>>> >>>>> with the >>>>>> >>>>>> concept that the ID should 128/160 bit integer? Is it possible >>>>> >>>>> that >>>>>> >>>>>> each peer in one domain assigned similar identity? Any >>>>> >>>>> suggestions? >>>>>> >>>>>> Thank you. >>>>>> >>>>>> Best Regards, >>>>>> Xianghan Zheng >>>>>> >>>>>> >>>>>> >>>>>> [2 xianghan_zheng.vcf <text/x-vcard; utf-8 (base64)>] >>>>>> begin:vcard >>>>>> fn:Xianghan Zheng >>>>>> n:Zheng;Xianghan >>>>>> org:;Information and Communication Technology >>>>>> adr:;;;Grimstad;;4879;Grimstad >>>>>> title:PHD Studnet >>>>>> tel;work:+47 3725 3441 >>>>>> tel;cell:+47 91664693 >>>>>> version:2.1 >>>>>> end:vcard >>>>>> >>>>>> [3 <text/plain; us-ascii (7bit)>] >>>>>> _______________________________________________ >>>>>> P2PSIP mailing list >>>>>> [email protected] >>>>>> https://www.ietf.org/mailman/listinfo/p2psip >>>>> >>>>> _______________________________________________ >>>>> P2PSIP mailing list >>>>> [email protected] >>>>> https://www.ietf.org/mailman/listinfo/p2psip >>>>> >>>> _______________________________________________ >>>> P2PSIP mailing list >>>> P2PSIP@ is well outside Alto'ietf.org/mailman/listinfo/p2psip >>> >>> > > _______________________________________________ > P2PSIP mailing list > [email protected] > https://www.ietf.org/mailman/listinfo/p2psip > -- Sincerely yours, Lichun Li Mobile Life New Media Lab, BUPT _______________________________________________ P2PSIP mailing list [email protected] https://www.ietf.org/mailman/listinfo/p2psip
