Performance modeling, queuing network models of computer systems and
networks, computer architecture and operating systems
Current research interests
I started my
research by designing and implementing a load simulator for the ESOPE Operating
System within the operating system research group at I(n)RIA, the French
national research center. Next, I
worked on modeling the performance of virtual memory systems, and, in
particular, on the determination of the optimal multiprogramming degree. During this work I started to build a
general framework for the equivalence and decomposition method which relies on
the use of marginal and conditional state probabilities. This method allows to obtain (in
general, approximate) solutions of complex systems through structured
decomposition.
During my tenure
at the Computer Science department at the ENST in Paris, I formed a research
team in adaptive system architectures.
Among the goals of the project was the dynamic adaptation of the
architecture to changes in load and to component failures. The resulting dynamic reconfiguration
was to allow the addition or removal of system components and devices without
the need to stop and reboot the system, well before "plug and play" systems
became a commercial reality.
Later, I joined a large computer manufacturer in the Silicon Valley (Amdahl Corporation) where I worked on the performance modeling of transaction processing systems, large enterprise systems, storage subsystems, as well as on modeling network performance. The latter resulted in my work (with L. Jow) on tandem queueing networks with blocking. My work in industry was instrumental in allowing me to develop an appreciation for the "real world", problems in model calibration, and, in general, the size of realistic problems.
As a professor of Computer
Engineering at the University of California Santa Cruz, my research topics have
included solution method for large systems, multi-server priority queues,
multicast systems with periodic synchronization and the modeling of optical
networks (during my sabbatical at Telecom Sud Paris). Additionally, I have continued my work
on model calibration. Some of the
ideas of that research led to the PhD work of Thomas Begin at the
University of Paris (UPMC), which incorporated the idea of automated model
calibration and selection.
At the same time, I have continued my work on
the use of conditional probabilities in the solution of queueing models. This idea resulted in the
development of numerically stable efficient recurrent solution of
M/Ph/1-like queues, and more recently, to the development of an
efficient iterative solution for the Ph/Ph/c queue. I continue to be interested
in approaches that allow circumventing the combinatorial explosion of the
size of the state space, in particular, in the context of virtualized systems
and "cloud computing".
PUBLICATIONS
?>
Journal Papers
Conference
Papers?>
Technical
reports
1.
Brandwajn,
A. and T. Begin, 2010. "A Recurrent Solution
of Ph/M/c/N-like and Ph/M/c-like Queues", INRIA
research report 7321, June 2010, Lyon, France.
2.
Brandwajn,
A. and S. Fdida. 1996. "Modeling and Analysis of a Transport Multicast
Protocol", UCSC-CRL-96-20, University of California, Computer Research
Laboratory, Santa Cruz, CA.
3.
Brandwajn,
A. 1993. "An Approach to the Decomposition of Large Stochastic Models",
UCSC-CRL-93-42, University of California, Computer Research Laboratory, Santa
Cruz, CA.
4.
Brandwajn,
A. 1987. "Aspects of Path Sharing in I/O", Technical Report, UCSC-CRL-87-16,
University of California, Computer Research Laboratory, Santa Cruz, CA.
5.
Brandwajn,
A. 1987. "A Simple Model of an Automated Library System", Technical Report,
UCSC-CRL-87-14, University of California, Computer Research Laboratory, Santa
Cruz, CA.
6.
Brandwajn,
A. 1986. "A Model of Cached I/O", Technical Report, UCSC-CRL-86-26, University
of California, Computer Research Laboratory, Santa Cruz, CA.
7.
Brandwajn,
A. and Y.L. Jow. 1986. "An Approximation Method for Tandem Queues with
Blocking", Technical Report, UCSC-CRL-86-19, University of California, Computer
Research Laboratory, Santa Cruz, CA.
8.
Brandwajn,
A. 1979. "Equivalence and Decomposition in Queuing Network Models - A Point of
View", Technical Report, CS-1979-6, Duke University, Durham, NC, May 1979.
9.
Brandwajn,
A. 1979. "A Method for Finding the Roots of Polynomials With Only Real Zeros",
Technical Report, CS-1979-2, Duke University, Durham, NC, February, 1979.
10. Brandwajn,
A. 1976. "A Finite-Source Queueing Model with Blocking", Technical Report
ENST-D-76008, Paris, France, June 1976.
11. Brandwajn,
A. 1976. "A Note on the Accuracy of the Equivalence and Decomposition Approach",
Technical Report ENST-D-76007, Paris, France, May 1976.
12. Brandwajn, A. 1975. "Equivalence et Decompostion dans les
Modèles à Files d'Attente et leur Application à
l'Evaluation des Performances de Systèmes d'Exploitation", Docteur d'Etat
Thesis, Paris, France, January 1975.
13. Brandwajn, A. 1972. "Procédure d'Exploitation pour une Liaison UC-UC",
Docteur-Ingénieur Thesis, Paris, France.
Invited
talks
Recent Sabbatical Leaves
April –June 2010
?>
I spent my Spring 2010
sabbatical at the LIP Laboratory of the Ecole Normale Supérieure de Lyon
continuing my joint work with Dr. Thomas Begin on efficient solutions of general
queues, as well as collaborating with colleagues (Professors Paolo Gonçalves and
Isabelle Guérin Lassous) and graduate students on topics related to modeling TCP
performance and on static versus dynamic network routing protocols. The work on these topics is still under
way.
During my stay in Lyon, I
gave a seminar on my current research interests. I was also the Chair of the doctoral
defense committee for Dinil Mon Divakaran (“Dealing
with Elephants in the Internet: Towards a Flow-Based Internet
Architecture”).
January-March
2011
I spent my Winter 2011 sabbatical in Paris dividing my time between the Ecole Nationale Supérieure des Télécommunications (ENST) and the LIP6 laboratory at the University Pierre et Marie Curie (Paris VI). My main research during this period was in the area of applicability of state sampling to the solution of models of large dimension (ENST with Dr. Thomas Bonald), as well as automated model generation and calibration (LIP6 with Prof. Serge Fdida). During my stay in Paris, I gave an invited talk at the LINCS Institute in Paris on possible approaches to cicumvent the intrinsic complexity of standard queueing network solutions.
The work on state sampling expands my previous work on fast solutions with partial state evaluation in computer and network models. The question addressed here was the applicability of this method to networks with strong coupling for which an exact product-form solution exists but is costly to compute. The preliminary results seem encouraging at least for some classes of models. The results obtained will become the subject of a future paper.
The
work on automatic model generation and calibration was an attempt to apply
methods developed previously during the Ph. D. work of Thomas Begin to the
modeling of the behavior of wireless networks. This work is still ongoing
Classes taught
recently