John Friedmann  -- AT&T WorldNet

Our interests in Web Caching are several fold. 

First, to improve  the overall WorldNet user experience by improving
the latency  to our end users of Web and other content. In this regard,
we are very interested in advanced caching algorithms such as prefetch
to improve the latency as much as possible. We are currently
undertaking a large scale data mining effort to capture data from
WorldNet which will help us to design better caching strategies and
test the effectiveness of various prefetch algorithms, as well as to
design and enable prefetch algorithms (since it is helpful to know what
to prefetch).  The specific data mining analysis that will most
faciliate this aspect is URL sequence analysis which  we are beginning
to examine through our data warehouse.

Second,  to examine the most efficient ways to route traffic across our
network including eliminating the need to route repetitive traffic
where possible. Web caching is very helpful in this regard and we are
very much interested in understanding the most effective distributive
caching techniques. We are currently developing simulation algorithms
to model distributed caching and the economic impact to help justify
the use of caching within WorldNet. In this regard, we want to
understand the benefits of Multicast  particularly in respect to Web
push technologies to more efficiently channel traffic across the
network.  For example if everyone accesses push channels on slightly
different schedules, then the network traffic load becomes
exponential.  It is as if everyone is watching the same tape on a VCR,
but noone is synchronized on the same frame. But if we can move to a
Cable TV or Digital Satellite model where we rebroadcast channels
every  hour, then we can efficiently use Multicast technology to route
Web push traffic across the network. This is as much a marketing
question as it is a technical issue, and it cannot be controlled solely
by ourselves, so we will be very interested to see where both industry
and the research community are heading.

Third, to ensure compliance with copyright , advertising and  content
freshness concerns of content publishers. We want to attempt to enhance
the business model for content publishers by providing an efficient
delivery channel for their information and provide adequate accounting
data for their advertisers, as well as ensure the freshness of the
data. In this regard, better dynamic content tagging mechanisms are
needed  as well as timeout parameters need to be truly synchronized
with the currency of the data. We are very interested to learn what
others are doing about these important questions.

Fourth,  there are many game theoretic aspects of Web caching  that are
interesting as both research problems and as possible business
strategies between competing networks, competing publishers and between
networks and publishers.  Some of the game theoretic problems are (1)
use of Web caching to disintermediate a competing network  who may be
directly linked to a content publisher, (2) to avoid the future payment
of settlements to peering networks,  (3) network strategies to force
content publishers to comply/cooperate with optimal network
distribution strategies of content  (e.g. see point two above), (4)
using  network based web caching as a generator of advertising revenue
and thereby potentially competing with content publishers.

Obviously, I have only briefly touched on these various issues and
would be happy to discuss my thinking on these various topics in more
detail at the conference.