Jim Gast's Selected Publications
7390 Computer Science & Statistics
University of Wisconsin / Madison
Madison, WI 53706
Office: (608) 262-6629
Lab: (608) 262-6003
Supported by the Anthony C. Klug NCR Fellowship in Computer Science
- Various on-line talks during CS-740. CS-740 is the
seminar course for Advanced Topics in Networking at University of Wisconsin / Madison.
I taught the course and led half of the discussions. Each student led one discussion.
The CS-740 talks web page contains pointers to the on-line talks,
including an introduction by me.
The Main CS-740 Web Page contains related information about the course
and the papers discussed.
Representing the Internet as a Succinct Forest
in IEEE Networks, March, 2004.
Combining previous techniques for inferring the structure of the Internet
ISP connections, this paper creates an acyclic tree that can be used for
Internet size and Internet growth estimation. A practical use for
solving cache placement problems is presented.
Available in PDF.
Topology Aware Estimation Methods for Internet Traffic Characteristics
Ph.D. Thesis, August, 2003.
Presents a model for analyzing aggregate traffic at one edge of Internet 2 to infer congestion in the backbone based on one-way delay experiments from the Surveyor project. The nature of harmonic congestion
propagation is explored and measurements are proposed that can be used
to inform future growth.
Available in PDF and
Cache Placement Methods Based on Client Demand Clustering.
This Technical Report (workshop version in HTML
or full version PS)
presents a method for optimal placement of proxy caches and
web services in the
Internet. Placing redundant services like Content Distribution Networks close to
clients improves performance from the client perspective. The paper uses information
from Border Gateway Protocol to derive a forest of trees that represent the major
branches of the Internet. Client demand from web server logs is aggregated and a dynamic
programming algorithm identifies optimal points in the trees for service placement.
Resource Deployment based on Autonomous System Clustering.
This conference paper has been accepted for Global Internet 2002, Taiwan, Nov. 2002. It
presents algorithms for annotating and augmenting an Internet Topology representation
to combine BGP tables with traceroute results. The resultant AS forest is a
compact representation of the Internet that has better prediction capabilities.
Available in PDF
- Internet Congestion Characteristics and Implications,
This document was my
prelim paper and
slides to present my research agenda to
my thesis committee. It poses the basic questions of tracking, graphing,
and ultimately understanding how congestion affects traffic flow in the
Internet. In particular, what does a congestion event look like? Does
congestion propagate to neighboring links? How can we visualize global
Internet congestion in a meaningful way in near-real-time? Since the
prelim document is a research proposal, it asks more questions than it
answers, but even the questions are thought-provoking.
How would you build a lab that could emulate the Internet backbone?
Even if you knew the parameters, how would you validate that the parameters
used in the model legitimately reflect the values seen in the real backbone?
Tape and Removable Storage
Disk and File Servers
Metrics to Use on the Road to Hierarchical Storage Management.
A paper presented at IEEE CompCon, 1995.
The paper presents a set of metrics
for managing and maintaining a storage farm that includes both high
and low latency storage. Simple metrics like "youngest file migrated" and
"bytes of files de-migrated" are presented and explained. The paper concludes
with ways to optimize future spending to tradeoff the utility of
adding more on-line storage versus the utility of adding more bandwidth and
capacity to robotically-available near-line capacity.
- Storage Networking Industry Association.
This introduction was the talk given at the formation meeting September 5, 1997
in Provo, Utah. After welcoming the first 6 companies, the talk introduces the
challenges and opportunities of an industry consortium based on moving proprietary
interfaces into an open market and creating open standards where new ones are needed.
Since then, the SNIA
has grown to over 300 member companies.
Impact of Distributed File Systems on Storage Management Services.
A talk presented at Network Storage '96, May, 1996.
This talk introduced the new concepts and APIs needed for backup, replication,
and archiving in a multi-architecture environment. Replicas of files need to
include extra metadata beyond simply the owner and permissions. Original files
need to know about their protection status, backup files need to know if their
original file has disappeared or changed, and replicas need to know how often
to synch up.
Managing Distributed Data on the Smart Global Network.
A talk presented at Brainshare 1996, Salt Lake City, April, 1996.
This talk was the first public peek at Novell's new NetWare Storage System
and the changes that would be needed to take full advantage of some
of the new features. This 64-bit file system erased limitations
of the prior file system. Many of the existing third-party backup and
archiving systems would automatically gain the benefits if they were correctly
using the Storage Management Services APIs.
- Developing SMS Backup. A tutorial presented at Interop 1994, Atlanta,
September, 1994. This tutorial introduced the concepts and APIs of Storage
Management Services, a cross-platform architecture for preserving and restoring
all of the data and metadata of files. Platform-dependent Target Service Agents
translated metadata into
System Independent Data Format. This work went on to become the first, new
ISO standard tape and optical disk format in over a decade (ECMA-208 and ISO/IEC-14863).
This tutorial also discussed the Universal Disk Format (UDF) which was later
extended by the Optical Storage Technology Association
and used as the disk format for all current DVDs.
- FileSys Presentation. Slides for my talk
on the difference between kernel-level file servers and user-level file servers.
- Common file sequence prediction. This paper examines the sequence of file
opens typically seen by programs like Microsoft Word. Dozens of DLLs and initialization
files are opened in a sequence that could be easily optimized. File systems that
recognize the onset of such sequences can accelerate the file opens in anticipation.
(hardcopy only, I have to re-type it?!?)
Talks on file replication via http to be added.
This page created and maintained by Jim Gast