Session III Plenary Notes

---

Pete Lega (PL)
C|Net: The Computer Network
``Virtual Software Library''

---

Notes by Raj Vaswani (@Home Network)

The talk

PL began with a history of the Virtual Software Library (VSL): originally running as a small- to medium-scale index on a single, Sparc 5 class workstation, it grew tenfold after moving to C|Net, where it is now hosted on a bank of Sparc 1000s. PL said that many companies (e.g., Adobe, IBM) are looking at the VSL as a one-stop software distribution source.

PL went on to describe features and ``non-features'' of what he termed ``popular meta-archiving''.

Features:

• One-stop retrieval.
• Acting as an ``agent'' for assisted search (resource discovery).
• Acting as an ``agent'' for locating best-bandwidth sites (VSL incorporates automated bandwidth testing against remote sites).
• User customization.
• Providing relevant editorial content (via tie-ins with C|Net's ``C|Net Online'' TV show.

Non-features:

• Content control. PL described a problem with having an automatically maintained ``Top N Downloads'' page, namely that some of those top downloads could be potentially offensive in nature.
• Local file caching. The VSL provides meta-indexing, not content hosting.
• Control over archiver-maintained data.
• Control over file delivery.

PL said that the VSL currently lists 50+ archives, indexing 50+ GB of data. He said that a great deal of human intervention (too much) was required to maintain VSL, pointing out that the proliferation of less computer literate users is increasing the amount of time spent in guiding archivers.

PL next described VSL Open Format 1.0, C|Net's convention for resource description. The major components of VSLOF 1.0 are:

• Archive Description Format (ADF), which allows an archive to register with the VSL system by specifying information about itself, e.g., its location.
• Quick Index File (QIF), a resource description format that seemed to encompass a fairly broad range of styles from something similar to an ``ls -l'' listing to free-form one-line descriptions. PL expressed a need for index enhancement tools, possibly using categories or keywords. He supported this by pointing out that a free-form description for, e.g., ``winzip'' may well not contain the word ``compression'', making it more difficult to find.

Finally, PL sketched requirements for VSL Open Format 2.0:

• Categories and keywords to improve searching.
• Per-file registration.
• Commerce and tech support enablers at the file level. E.g., notation for stating that a given version of the file replaces some list of previous version.
• Even more file-level information (e.g., how to get in touch with the software's author).
• Compatibility with VSLOF 1.0.

Q&A

In the interest of maximizing breakout session time, Q&A following talks had been eliminated.

---

Luis Gravano (LG)
Stanford University
``Informal Internet Standards at Stanford''

---

The talk

LG stated the goal of the Stanford work was to coordinate agreement between search engines, content providers, and users. Specifically, they aimed to facilitate 3 tasks:

• Resource discovery (which engines should respond to a given query).
• Query formatting (how to format queries for submission to engines).
• Result merging (how to present a unified set of results to users).

LG felt that standardization as occurred with SQL -- rapid common agreement -- was rare, and enumerated what he felt were 4 more common approaches to standardization:

• De facto or ``brute force'' standardization, e.g., imposed by a dominant vendor. LG argued that such ``standards'' may be poorly designed, and that discussion of improvements is typically irrelevant.
• Early-deployment standardization, e.g., VSLOF. LG argued that such ``standards'' may be poorly designed due to hasty decision making, or may not keep pace with a changing environment. He felt that the ``giants'' may appropriate such ``standards'' anyway, citing HTML/Netscape as an example.
• No standardization, with dynamic negotiation compensating for the absence of standards. LG felt that the benefits of this approach were that it allowed for competing solutions, did not inhibit progress, and avoided market surprises (i.e., was flexible in the face of environmental changes). He felt that the approach did have some drawbacks, arguing that it had: high maintenance cost, increased latency at runtime, complicated implementation, and limited power.
• Some ``lightweight'' standardization agreement, with dynamic negotiation. LG felt that this was the best compromise between individuality and interoperability, since companies could continue to innovate within a cooperative framework, with features enjoying significant commonality migrating, over time, into a more rigid standard.

LG urged the community first to agree on minimum functionality, and then to add runtime negotiation. He listed some areas requiring attention:

• Specifying minimum common functionality (e.g., for search engines, the query language; e.g., for content providers, the metadata exported).
• Specifying ways to describe heterogeneity (e.g., for search engines, the manner in which a given engine differs from another; e.g., for content providers, resource descriptions).
• Specifying ways to deal with heterogeneity (e.g., converting relevance rankings from various engines' proprietary algorithms to provide a unified ranking).

Q&A

In the interest of maximizing breakout session time, Q&A following talks had been eliminated.