France Master Internship PARIS

September 14, 2007

Sponsored Links

Career Blog

Online Distance Learning Courses for International Students

Google Internships for International Students

Master Internship Proposal: Hard Drive Input/Output Monitoring in Clusters / Study of the Influence of I/O on Node Performance

Location: IRISA/INRIA Rennes (France)
Team: PARIS (http://www.irisa.fr/paris)
Duration: 4 to 6 months

Start date: ASAP
Contact and application: Christine.Morin at irisa.fr, Adrien.Lebre at irisa.fr

Context
Since June 2006, the PARIS project-team at IRISA (INRIA research center in Rennes) has been involved in the XtreemOS European project (coordinated by Christine Morin, senior researcher in the PARIS project-team). This 4-year project aims at designing, implementing and evaluating a Linux based operating system (OS) for Grid architectures (http://www.xtreemos.eu).

The XtreemOS Grid OS will be executed on different kinds of computers in a Grid. There will be three XtreemOS flavours, one for each kind of Grid node: individual computers (typically for PCs), clusters, and mobile devices.

The PARIS project-team is involved in the design and implementation of the cluster flavour of XtreemOS. The cluster flavour of XtreemOS is based on a Linux Single System Image (SSI) operating system for clusters that leverages the Kerrighed technology originally developed in the PARIS project-team (http://www.kerrighed.org). A Linux SSI operating system provides the illusion that a cluster is a virtual multiprocessor machine executing Linux. The SSI concept is implemented by a set of distributed services for global resource (processor, disk, memory) management. For instance, Kerrighed implements KDDM to allow data sharing in the cluster, process migration for load-balancing, checkpointing mechanisms, a distributed file system (kDFS)…

Work to be done
In this context, we offer a Master internship position (4 to 6 months) in the PARIS project team at the INRIA/IRISA laboratory located in Rennes (France).
The subject focuses on the influence of data accesses (hard drive I/O) on node performance.
In order to improve the use of a cluster, schedulers have to take into account resource usage by each process. A lot of work has been done on schedulers that do not only take into account the processor load such as for example in the MOSIX operating system [BAR98]. However, these previous studies consider CPU and memory monitoring mechanisms without taking into account I/O aspects.

The main objective of this internship is to provide the missing I/O probes capable of notifying schedulers about current concurrency within a node. The work will be divided into two parts:

Analysis and I/O Model
After a study of the related literature, the student will have to analyse how processes impact on each other within a node when hard drives are stressed (network intensive process versus I/O intensive process, CPU intensive versus I/O intensive, memory intensive versus I/O intensive, number of processes, ….). Moreover taking into account “user” processes load, he/she will have to consider resource usage exploited by the SSI file system itself [BECK06].
These experiments should lead to a first I/O model needed for the second part of the internship.

Design and Implementation of I/O probes
After analyzing concurrency between user and kernel processes and based on the resulting model, the student will design and implement the I/O probes in the context of Kerrighed SSI that provides a scheduling framework facilitating integration of new probes and customized scheduling policies. At the first sight, I/O probes development will be done at kernel level for the major part. Experimentations will be carried out on a cluster of the Grid’5000 experimental platform (http://www.grid5000.org).

The student should be skilled enough in operating system (Linux), distributed system and
system programming (C language).

http://www.kerrighed.org
http://www.xtreemos.eu

[BAR98] A. Barak et O. La’adan. The MOSIX multicomputer operating system for High
Performance Cluster Computing. Future Generation Computer Systems, 1998
[BECK06] Beckman, P., Iskra, K., Yoshii, K., Coghlan, S., The Influence of Operating
Systems on the Performance of Collective Operations at Extreme Scale Cluster Computing,
IEEE International Conference on Cluster Computing, 2006

Full Position Details Here

Scholarship