ObjectWeb Consortium
Search ObjectWeb Mail Archive: 

Advanced Search - Powered by Google


Mail Archive Home | community List | October 2005 Index

<--  Date Index     <--  Thread Index    

post-doctoral position


A post-doctoral position to contribute to ObjectWeb projects CLIF and Fractal...

Performance evaluation in autonomic software infrastructures for self-optimization

General Information

• Contract Type: Post doctorate position for a limited duration of 12 or 18 months
• Starting date: Early 2006
• Domains: autonomic computing, performance evaluation and tuning
• Institution: France Telecom, R&D Division, MAPS/AMS lab, 28 chemin du vieux chene, BP 98 38243 MEYLAN cedex, FRANCE
• Locality: Grenoble agglomeration, France
• Contact: Bruno Dillenseger, ph. +33 4 76 76 41 46, bruno.replace_with_my_last_name@francetelecom.com

Context

The post-doctoral fellow work will contribute to an internal pluri-disciplinary research project on Autonomic Information Networking. The contribution of the host team is about developing some elements of an autonomic software infrastructure. Autonomic computing (according to IBM terminology [1]) aims to reduce the complexity of managing software systems. To be autonomic, a system must configure and reconfigure itself, continually optimize itself, recover from malfunction, or protect itself, while keeping its complexity hidden from the user. A minimal autonomic software infrastructure would typically include:
• a framework allowing the realization of censors and actuators on the system being managed;
• an event transport mechanism;
• a control loop (feedback) mechanism.
More specifically, the proposed work will focus on performance evaluation and self-optimization issues, reusing, completing and adapting the autonomic software infrastructure developed by the team.

Problem

In the context of autonomic computing systems, a self-optimizing system has to be able to evaluate its own performance. To achieve this evaluation, the system must observe itself via some probes and analyse measurements produced by those probes. Then, decisions must be made and actions must be taken (reconfiguration) so as to improve performance, thus implementing a control loop.
Among these steps, one key issue is the way those data can be analyzed, and how conclusions may be drawn from this analysis. Gathered measurements are:
• computing resources usage:
o at hardware or operating system level (e.g. memory occupation, processor time, context switch rate, network bandwidth, disk transfer rate, etc.);
o at arbitrary software levels (e.g. data base cache efficiency, request waiting queue length, size of a thread pool or whatever specific to whatever software element);
• system performance itself (request processing time and throughput, error/rejection rate, latency, etc.)
As of today's practice, this kind of analysis typically involves a lot of empirical know-how and manual tasks, with a mix of heterogeneous tools (ad hoc scripts and programmes, mathematical and plotting applications, spreadsheets, etc.).

Subject description

In a first step, this work will mainly consist in designing and prototyping relevant analysis tools. Practical experiments will be performed using CLIF [2] load injection (traffic generation) framework,
including its probe layer, and an arbitrary sample system under test to be chosen (for instance, a web application).
Following the same example, the second step will consist in making proposals to introduce self-optimization on both the load injection side and the system under test side. In other words, analysis tools will be embedded and hidden so as to make autonomous reconfiguration decisions without human interaction. Indeed, the load injection part must be observed and optimized in order to actually and properly generate the specified traffic (e.g. by setting the right size of a thread pool).
Prototypes will be implemented in Java. On-going work in the hosting team may provide support for:
• decision making (use of Event-Condition-Action rules),
• distributed, component-oriented architecture based on the Fractal model [3],
• JMX [4] and Fractal based instrumentation and communication middleware,
• experimental, instrumented, Fractal-based web server.

Qualification

We are looking for a candidate with a "software profile", typically a PhD in computer science with work in a relevant field: distributed computing and middleware, performance evaluation, dynamic adaptation, mathematical and statistical analysis, component-based systems.
The candidate should be a good Java programmer and feature ability to work in team, as well as to seek and propose solutions and new ideas. The candidate should also be keen on open source software.

Environment

The candidate will work within the France Telecom R&D Division MAPS/AMS laboratory specialized in distributed middleware architecture and technologies, and specifically within the research unit POLAIR (Distributed Software Architectures and Infrastructures). The work will take place at France Telecom R&D division premises in Meylan (Grenoble agglomeration), France.

References

[1] IBM Research, Autonomic Computing: http://www.research.ibm.com/autonomic/
[2] CLIF is a Load Injection Framework: http://clif.objectweb.org/
[3] Fractal component model: http://fractal.objectweb.org/
[4] JMX, Java Management eXtension: http://java.sun.com/products/JavaManagement/

<--  Date Index     <--  Thread Index    

Reply via email to:

Powered by MHonArc.

Copyright © 1999-2005, ObjectWeb Consortium | contact | webmaster.