Reference persistence implies (a) that references should not be tied to particular administrative domains or entities, as they currently are in DNS and (b) that references should be at the level of objects rather than entire sites. If someone creates a Web page at one institution then changes his affiliation or network provider, then maintaining persistence would require that the first institution continue to serve the individual's Web page (or provide an HTTP redirect) for all time, which is an impractical expectation . Also, without resorting to HTTP redirects, the current Web infrastructure has no way for individual objects to separate from their sites and migrate cleanly: today, if an object moves, everyone linking to the object must update their references.
DNS has become a branding mechanism and, as a result, fights over domain ownership are common. Social problems include name squatting, typo squatting, and lawsuits over trademark infringement. Although disputes over human-readable names are inevitable, we believe the reference resolution infrastructure is a poor place to resolve those disputes. In fact, we believe the infrastructure should force references to be inherently human-unfriendly. Of course users must be able to associate meaning to references, but the binding between opaque references and human-friendly names should be done outside the referencing infrastructure. Such a separation would (a) free the RRS to focus only on technical concerns and (b) permit multiple, competing solutions to human-friendly naming.
Unlike DNS-based URLs, flat, semantic-free references have no explicit structure to give resolution hints. Until recently, there was no way to resolve references scalably in such a namespace, which is largely why the URN literature chose to use a partitioned set of context-specific resolvers. However, the recently developed Distributed Hash Table (DHT) technology is exactly designed to map from an unstructured key to a network location responsible for the key. SFR uses DHTs to map each object reference to a machine that contains object meta-data, such as the object's current IP address and the pathname of the object. Once an application, such as a Web browser, has this meta-data, it can retrieve the actual object.
We imagine that SFR would be deployed on a managed infrastructure (as mentioned above, we believe the namespace should be unmanaged; the infrastructure, however, is a different story), not on the desktops of random cable modem users. To repeat: even though SFR uses DHTs, which are a so-called peer-to-peer technology, we are not relying on flaky personal machines connected via cable modems!
SFR's features naturally do not come without cost since many of the desirable features of today's Web derive from DNS. As examples, DNS's hierarchical structure enforces URLs' uniqueness and provides fate sharing (a disconnected institution can still access local pages) while the human readability of DNS hostnames gives users some (perhaps misguided) confidence they have reached their desired data. Some of the challenges SFR must address include:
 Besides the URN literature, similar observations have also been made
by: (a) Michael O'Donnell, in his Proposal
to separate Internet handles from names; (b) Bob Frankston in this
essay; and (c) the Globe
Project (see especially the paper Locating objects in wide-area systems, IEEE Communications Magazine, January 1998; here is
ps or pdf).
 The current approach, in which individuals maintain Web pages with domains like www.personalname.org, does not allow an individual Web object (such as a Web page or a directory of photographs) to separate from its original site. Also, even today, this approach might be awkward or inappropriate in contexts when content should not be named by a particular individual.
The following papers give more information about SFR.
The first is a position paper for a workshop; it outlines an earlier
version of our philosophy.
The second is a report for a student workshop and contains a later
version of our philosophy.
The third is a full-length conference paper and contains the most refined
statement of our philosophy along with design and implementation details.
Hari Balakrishnan, Scott Shenker, and Michael Walfish
2nd International Workshop on Peer-to-Peer Systems (IPTPS '03), Berkeley, CA, February 2003.
[PostScript (61KB)] [Gzipped PostScript (25KB)] [PDF (62KB)]
First IRIS Student Workshop, Cambridge, MA, August 2003.
[PostScript (100 KB)] [Gzipped PostScript (34 KB)] [PDF (91 KB)]
Michael Walfish, Hari Balakrishnan, and Scott Shenker
1st Symposium on Networked Systems Design and Implementation (NSDI '04), San Francisco, CA, March 2004.
[PostScript (615KB)] [Gzipped PostScript (130KB)] [PDF (142KB) ]
We plan to release a prototype of SFR shortly; this prototype is layered on top of the Chord and DHash system.
M. I. T. Laboratory for Computer Science · 200 Technology Square · Cambridge, MA 02139 · USA