The SEER Predictive Hoarding System

SEER is a predictive hoarding system for disconnected mobile operation. Briefly, it observes the user's file-access patterns while a mobile computer is connected to the network, and uses those to predict what files will be needed when the network is no longer available. It then arranges to have those files stored on the mobile machine in advance of disconnection.

SEER is part of the larger Travler research project, which is developing system support for the mobile and nomadic computing that will be common in 21st-century computers.

As part of the SEER project, traces of user behavior were collected over several months. These traces are now available to other researchers.

SEER Overview

(Figure showing a block diagram of SEER)

SEER observes the user's references to files and feeds these to a special correlator process. The correlator uses this information to infer relationships between various pairs of files, and encodes these as distance estimates. In addition, external investigators can provide information about inter-file relationships by inspecting things like file contents.

Periodically, the correlator integrates all of the information it has about various files, and uses this information to generate clusters of files that represent projects. The most important clusters are identified, and this information is passed to a replication substrate such as Rumor, which will store copies of these files on the mobile machine so that they will be available at disconnection. user.

Publications related to SEER:

G. H. Kuenning and G. J. Popek. Automated Hoarding for Mobile Computers. Proceedings of the 16th ACM Symposium on Operating Systems Principles, (SOSP-16) St. Malo, France, October 5-8, 1997. Available in either Postscript or HTML format. This is a detailed discussion of Seer and its construction. For persons interested in how Seer works, but who don't have the time to read the dissertation, this is the best choice.

G. H. Kuenning, P. Reiher, and G. J. Popek. Experience with an Automated Hoarding System Personal Technologies 1, 3, September 1997. This paper gives many tables and figures that document the performance and behavior of Seer. Some of the material is taken from the Seer dissertation, but other material appears only here.

G. H. Kuenning, The Design of the SEER Predictive Caching System, Proceedings of Mobile Computing Systems and Applications 1994, Santa Cruz, CA, December 1994. Although some of the ideas in this paper are outdated, this is still the best overview of Seer for people who just want a quick idea of what's going on.

G. H. Kuenning, G. J. Popek, P. Reiher, An Analysis of Trace Data for Predictive File Caching in Mobile Computing, Proceedings of the 1994 Summer Usenix Conference. This paper appears in substantially the same form as Chapter 2 of the Seer dissertation, referenced below.

G. H. Kuenning. Seer: Predictive File Hoarding for Disconnected Mobile Operation. Ph.D. dissertation. Available either in the original single-sided, double-spaced form (301 pages total) or paper-saving double-sided, single-spaced form (174 pages) as Technical Report UCLA-CSD-970015. This is the definitive document on Seer, straight from the horse's mouth.

Geoff Kuenning's home page.

This page maintained by

Geoff Kuenning.