Post-Docs in Luxembourg University

Les annonces, les colloques, les postes, ...

Modérateur : kaiser

Répondre
kaiser
Messages : 323
Inscription : 03 nov. 2004 17:40

Post-Docs in Luxembourg University

Message par kaiser » 11 août 2008 15:32

Date : 10 août 2008 21:08:54 HAEC
De : pascal.bouvry@uni.lu
Objet : [asr-forum] Post-Docs/PhD Students in Luxembourg
À : asr-forum@cines.fr

TrustIT

Different positions at Postdoc or PhD levels are available at Luxembourg University in the team of Prof Pascal Bouvry:
http://csc.uni.lu

Funding will be done through FNR (www.fnr.lu) AFR grants.
PhD positions are 3 years long and post-doc ones are 2 years long.

Profiles:

PhD and/or Master degree or equivalent in Computer Science, Computer Engineering or Applied Mathematics.
Working languages: English

Applications should be sent to the following address
pascal.bouvry<at>uni.lu


== Position 1

The Research Project will be carried out in co-operation with the
North Dakota State University, Fargo, ND, USA and the topic is Power
and Energy Efficient Resource Allocation in Large-scale Data Centers.

Project description:

Around the globe, data centers are running into a power and cooling
barrier as they try to deploy new servers, storage and blades, which,
in addition to consuming more power than their predecessors, are often
more dense. In essence, if IT managers cannot find better ways to take
the heat out of data centers, they cannot take advantage of ongoing
improvements in processor performance and system density. Of course
there is a flip side to the coin ñ Money. Energy costs are rising
sharply, new computing systems require more power, and data center
cooling systems are using more electricity. Some enterprises now spend
more on data center utility bills than they spend to acquire new
computers.

The main objective of this work is to explore possible areas and
investigate novel techniques for power and energy conservation of
server clusters that support huge data centers of today. Some of the
major objectives of this work are cited below.

1. Explore possible power saving areas.
2. Reduce total cost of the data center by operating within a reduced
power budget.
3. Develop Quality of Service (QoS) aware power management techniques.
4. Meet QoS while minimizing power utilization.
5. Develop power aware resource aggregation and resource partitioning/
scheduling for dynamic platforms to support data centers.

== Position 2

The Research Project will be carried out in co-operation with the
Colorado State University, CO, USA and the topic is Robust Scheduling on Desktop Grids.


“Desktop Grids” are based on volunteer computing: idle cycles of desktop PCs and workstations shared by the users worldwide are stolen through the Internet to compute parts of a huge problem. One of the best known project that run on such computing platform is SETI@Home which has now moved to BOINC. This platform is currently the most powerful computing system as a peak speed of 1.06 PetaFlops has been achieved in January, 2008.

The scheduling problem will be studied in two different manners. First, the static case will be studied, in which the load balancing will be made in an off-line way. After that, we will extend the problem to the complex case in which the scheduler has to deal with new arriving tasks dynamically (i.e., an on-line scheduler). In this second version of the problem, the algorithm must be fast and accurate enough to react to the new situations after new arriving tasks and different resource availabilities. In the two studies, realistic-size instances will be studied (considering grids of several thousands of heterogeneous nodes), much larger than the currently existing ones in the literature.

Ensuring a robust scheduling is a priority in this work. Here robust means that the probability that the resources finish the assigned tasks is maximized, and thus the system is not overcome in the case that some tasks are delayed. For this, it can be useful measuring a trust level on resources by quantifying the certainty that the resource will finish the job. Additionally, it could be possible to extend this approach to security issues, where the trust level would be a measure of the security we can expect from the machines.

Various objectives might be considered in terms of quality of the schedule. However we restrict ourselves in the context of the current work to minimization of makespan and maximization of the robustness. Hence, the problem can be studied as mono-objective optimization (maximizing robustness satisfying a makespan constraint, or minimizing makespan satisfying robustness constaint, or a using a weighted function of both makespan and robustness), or as a bi-objective optimization (minimizing makespan and maximizing robustness).

== Position 3

The Research Project will be carried out in co-operation with INRIA & the
Colorado State University, CO, USA and the topic is Ensuring confidentiality on desktop grids.


“Desktop Grids” are based on volunteer computing: idle cycles of desktop PCs and workstations shared by the users worldwide are stolen through the Internet to compute parts of a huge problem. One of the best known project that run on such computing platform is SETI@Home which has now moved to BOINC. This platform is currently the most powerful computing system as a peak speed of 1.06 PetaFlops has been achieved in January, 2008.
Confidentiality to protect the data exchanged together with the program executed from an external attackers interested to access such information. At this level, Software obfuscation is a software-only partial answer to software protection and data protection [11]. The code is transformed at the source or binary level to render more complex its readability and also change the data coding. Of course, often this slows down the execution and it is not impossible for someone very motivated to reverse-engineer the obfuscating process since the code is available for execution and can be exercised at will.
Another way is to transform an algorithm that produces data from input data into an isomorphic one that acts on ciphered input and produces ciphered output. It is also possible to add some authentication mechanisms in it to certify the computation done. For some simple algorithms there exist such efficient isomorphic algorithms but, unfortunately, this seducing approach has an intractable complexity for real life programs. Since basic computations (as, for example, floating point operations that are highly optimized in modern processors), are transformed in elementary operations executed to emulate some enciphered circuits, the expected efficiency on grids is no longer possible. Further effort are therefore required to find novel ways to deal with this issue and it will be one of the academic interest of this project;

Répondre

Qui est en ligne ?

Utilisateurs parcourant ce forum : Aucun utilisateur inscrit et 1 invité