SEMINAIRES CERIA

Université Paris-Dauphine
Place du Maréchal de Lattre de Tassigny - Paris 16éme
Métro: Porte Dauphine - RER Ligne C: Avenue Foch -
BUS PC: Porte Dauphine

Jeudi 11 mai 2000, 14h00

Salle A 709, 7ème étage, Bat. A ( Nouvelle aile du bâtiment de l'université)

" The Case for Efficient File Access Pattern Modeling" ,

Dr. Darrell Long and Thomas Kroeger

Dr. Darrel Long is Professor of Computer Science and Associate Dean of the
Jack Baskin School of Engineering at the University of California, Santa Cruz. He is also Visiting
Scientist at the IBM Almaden Research Center. His current research interests are primarily in high
performance I/O systems and video-on-demand. The Case for Efficient File Access Pattern
Modeling.

Résumé de la présentation:
Most modern I/O systems treat each file access independently. However, events in a computer
system are driven by programs. Thus, accesses to files occur in consistent patterns and are by no
means independent. The result is that modern I/O systems ignore useful information.
Using traces of file system activity we show that file accesses are strongly correlated with preceding
accesses. In fact, a simple last-successor model (one that predicts each file access will be followed
by the same file that followed the last time it was accessed) successfully predicted the next file 72%
of the time. We examine the ability of two previously proposed models for file access prediction in
comparison to this baselinemodel and see a stark contrast in accuracy and high overheads in state
space. We then enhance one of these models to address the issues of model space requirements.
This new model is able to improve an additional 10% on the accuracy of the last-successor model,
while working within a state space that is within a constant factor (relative to the number of files) of
the last-successor model.
While this work was motivated by the use of file relationships for I/O prefetching, information
regarding the likelihood of file access patterns has several other uses such as disk layout and file
clustering for disconnected operation.

Email address: Darrell Long <darrell@cse.ucsc.edu>
Organisateur: Professeur Witold Litwin
Mél: Witold.Litwin@dauphine.fr

Séminaire CERIA : Le  07 mars 2000, 13h00

Salle A 701, 7ème étage, Bat. A ( Nouvelle aile du
bâtiment de l'université)

Université Paris-Dauphine
Place du Maréchal de Lattre de Tassigny - Paris 16éme
Métro: Porte Dauphine - RER Ligne C: Avenue Foch -
BUS PC: Porte Dauphine

13h00 : Présentation des travaux de l'équipe du Professeur Henri BASSON (équipe MoDEL) du Laboratoire d'Informatique du Littoral (LIL) de l'université du Littoral.

14h00 : Présentation des travaux de l'équipe du professeur Witold LITWIN du Centre d'Etudes et de Recherche en Informatique Appliquée (CERIA) de l'université Paris-Dauphine.

****************************************

Séminaire CERIA : Le  29 novembre 1999 , 14h30

Salle A 709, 7ème étage, Bat. A ( Nouvelle aile du
bâtiment de l'université)

Université Paris-Dauphine
Place du Maréchal de Lattre de Tassigny - Paris 16éme
Métro: Porte Dauphine - RER Ligne C: Avenue Foch -
BUS PC: Porte Dauphine

"A Basis for Computer Understanding: A Classification and Performance Factors"

Dr. Tuncer I. Ören

Professor Emeritus, University of Ottawa, Canada and Marmara
Research Center,
Information Technologies Research Institute, Turkey

Email: tuncer@mam.gov.tr

Résumé: Understanding is one of the important philosophical topics. From a pragmatic point of view, it has a broad application potential in manycomputerized studies including program understanding, machine vision, fault detection based on machine vision as well as situation assessment. Therefore, systematic studies of the elements, structures, architectures, and scope of applications of computerized understanding systems as well as the characteristics of the results (or products) of understanding processes are warranted. Some basic concepts and definitions of understanding are
reviewed with appropriate references to their philosophical roots. A taxonomy of over 25 types of understanding is offered. The elements as well as the factors affecting performance of understanding systems are explained.

Présentation (fichier Powerpoint 7)

Economie des Nouvelles Technologies

Dr. Michel Volle

Directeur de Volle SARL,
Membre de l'Institut International de Statistiques,
Administrateur (hors classe) de l'INSEE

Email: michel@volle.com

Résumé: Le " système technique contemporain " est fondé sur la synergie entre micro-électronique, informatique et automatisation. On peut styliser sa fonction de production en supposant qu’elle est " à coûts fixes " : le coût de production indépendant du volume produit, est payé dès l’investissement initial.
Développons cette hypothèse : les usines étant des automates, l’emploi réside dans la conception et la distribution. La distribution des revenus n’est pas reliée au salariat. Les entreprises différencient leur production pour construire des niches de monopole. Elles organisent leurs processus autour du système d’information. Le commerce passe par des médiations empruntant la communication électronique. L’investissement est risqué, la concurrence est mondiale et violente. On retrouve dans cette présentation stylisée des aspects tendanciels de notre économie. Elle éclaire la description des secteurs de l’informatique, de l’audiovisuel, des réseaux (télécommunications, transport aérien, etc.), du commerce, ainsi que les aspects stratégiques et tactiques des jeux de concurrence dans ces secteurs et dans ceux qui les utilisent. On voit alors que cette économie hautement efficace pourrait aller au désastre si elle était traitée sur le mode du " laissez faire ", sans considérer lesexigences de l’éthique et de la cohésion sociale.

Disk Array Declustering with PDDL

Dr. Thomas Schwarz, S.J. University of California at Berkeley  

Disk arrays (a.k.a. RAIDs) combine high availability, high throughput, and cost effectiveness.  They achieve this by grouping stripe units (single disk blocks or groups of disk blocks) into stripes and by additionally storing the parity of the stripe units somewhere else in the disk array.  In order to maintain good performance in the case of a single disk failure, stripes should not be so large that they store a stripe unit on each disk.  These disk arrays are called declustered.  Good lay-outs for declustered disk array remain an open research problem with potentially large practical implications given the abundance of commercial raid-systems.  I will discuss criteria for optimal layout (given by Holland / Gibson) as well as some solutions.  I will present in particular a new layout scheme, permutation development data layout (PDDL), which uses the combinatorics of near resolvable block designs to generate good data layouts.  Finally, I will comment on the results of performance experiments and their consequences for the optimality conditions.  The talk reflects common work with Walter Burkhard and Jesse Steinberg from UCSD.

Hote :  Witold LITWIN

Heure : 16 h
Lieu :   Salle  A709.
            Univ. Paris IX Dauphine.
             Place du Maréchal de Lattre de Tassigny
             75 775 Paris Cédex 16
             Standard : 01.44.05.44.05
             Métro : Porte Dauphine - RER C : Avenue Foch

Pour toute information complémentaire : 
 
 

Efficient Distributed Backup with Delta Compression

Randal C. Burns,   Darrell D. E. Long
  Department of Computer Science
IBM Almaden Research Center,

University of California, Santa Cruz

Inexpensive storage and more powerful processors have resulted in a proliferation of data that needs to be reliably backed up. Network resource limitations make it increasingly difficult to backup a distributed file system on a nightly or even weekly basis. By using delta compression algorithms, which minimally encode a version of a file using only the bytes that have changed, a backup system can compress the data sent to a server. With the delta backup technique, we can achieve significant savings in network transmission time over previous techniques.
Our measurements indicate that file system data may, on average, be compressed to within 10% of its original size with this method and that approximately 45% of all changed files have also been backed up in the previous week.
Based on our measurements, we conclude that a small file store on the client that contains copies of previously backed up files can be used to retain versions in order to generate delta files.

To reduce the load on the backup server, we implement a modified version storage   architecture, version jumping, that allows us to restore delta encoded file versions with at most two accesses to
tertiary storage. This minimizes server workload and network transmission time on file restore.

==================================================

Improving Video-on-Demand Server Efficiency
Through Stream Tapping

Steven Carter  and  Darrell D. E. Long
Departement of Computer Science
Jack Baskin School of Engineering
University of California, Santa Cruz

Efficiency is essential for Video-on-Demand (VOD) to be successful. Conventional VOD servers are inefficient; they dedicate a disk stream for each client, quickly using up all available streams. However, several systems have been proposed that allow clients to share streams. We present a new system called stream tapping that allows a client to greedily ``tap'' data from any stream on
the VOD server containing video data the client can use. This is accomplished through the use of a small buffer on the client set-top  box and requires less than 20% of the disk bandwidth used by
conventional systems for popular videos. We present a description and analysis of   the stream tapping system as well as comparisons between it and other   efficiency-improving systems.

Pour toute information complémentaire :