From optical-networks@mailman.isi.edu  Fri Sep 27 23:29:22 2002
From: optical-networks@mailman.isi.edu (Joe Touch)
Date: Fri, 27 Sep 2002 15:29:22 -0700
Subject: [optical-networks] [Fwd: Fwd: Defense Announcement]
Message-ID: <3D94DBC2.40109@isi.edu>

This is a multi-part message in MIME format.
--------------090209020206070404020208
Content-Type: text/plain; charset=us-ascii; format=flowed
Content-Transfer-Encoding: 7bit

FYI - public is welcome to attend, in case you're wondering how we spend 
some of the rest of our time ;-)

Joe

--------------090209020206070404020208
Content-Type: message/rfc822;
 name="Fwd: Defense Announcement"
Content-Transfer-Encoding: 7bit
Content-Disposition: inline;
 filename="Fwd: Defense Announcement"

Received: from tap.isi.edu (tap.isi.edu [128.9.160.151])
	by boreas.isi.edu (8.11.6/8.11.2) with ESMTP id g8RMRfC21079
	for <touch@isi.edu>; Fri, 27 Sep 2002 15:27:41 -0700 (PDT)
Message-Id: <5.0.0.25.2.20020927151419.0271f698@boreas.isi.edu>
X-Sender: ahughes@boreas.isi.edu (Unverified)
X-Mailer: QUALCOMM Windows Eudora Version 5.0
Date: Fri, 27 Sep 2002 15:17:21 -0700
To: touch@ISI.EDU
From: "Amy S. Hughes" <ahughes@ISI.EDU>
Subject: Fwd: Defense Announcement
Mime-Version: 1.0
Content-Type: text/plain; charset="us-ascii"; format=flowed
X-AntiVirus: scanned by AMaViS 0.2.1

You wanted a copy to send to people...

--------------------

WHEN: September 30, 2002; 10:30am

WHERE: SAL 322

CANDIDATE: Amy S. Hughes

COMMITTEE:
Joe Touch (chair)
Christos Papadopolous
Ahmed Helmy (outside member)

TITLE: Enhancing Network Object Caches Through Cross-Domain Prediction

ABSTRACT:
Interactive applications must minimize user-perceived latency. Performance 
gains in all areas of computer software and hardware significantly reduce 
delays that users experience. Anticipating actions and exploiting resources 
removes delay from the critical path. In networked applications, increasing 
resources such as bandwidth eliminates most delays, but propagation delay 
is the result of the physical distances that separate clients and servers. 
When it is not possible to move resources closer to the user, the only way 
to avoid distance-based delays is to anticipate requests and satisfy them 
before the user makes them.

Anticipating requests requires a predictor function to generate likely 
future requests. Predictors exploit structure in the application, such as 
usage patterns or object characteristics. A key contribution of this work 
is a model for representing predictors and their effects on an application. 
The model is extended to represent predictors that involve multiple 
application domains. The extension of this model, cross-domain prediction, 
provides new opportunities for prediction in applications that lack 
internal structure or where that structure is insufficient to predict all 
requests.

Cross-domain prediction can reduce delays in the web by predicting requests 
that do not originate in the browser, such as URLs from e-mail messages. A 
test system comprised of a browser, a web proxy, and e-mail filters 
collected user interaction statistics about URLs embedded in e-mail 
messages. The results indicate that e-mailed URLs account for about a 
quarter of non-predictable requests. More than half of these URLs are 
static and prefetching these objects does not significantly increase 
request traffic or storage requirements. Even more opportunities for 
cross-domain prediction exist in other systems.

--------------090209020206070404020208--