Am I reliable ?

This week, I am attending SOSP !

I will present a paper at the HotDep workshop. This work was done with my former master student Huynh Tu Dang during its internship. The paper is about the reliability of placement algorithm when they perform reconfigurations.

Basically, it shows that we suck (well, or at least I suck) at developing placement constraints because we usually develop constraints that restrict the placement at the end of the reconfiguration process. This leads to temporary failures. The good point is that using BtrPlace, it is easy to implement continuous constraints that ensure there will be no violations, event during the reconfiguration.

As usual, the paper is available on my website. The slides of the talk will be available shortly.

Next post will be about the european project DC4Cities that started in september !

Btrplace fundaments: the no longer cursed paper

I am pushing for the practical interests of flexible VM manager since about 5 years now. In fact, I have started to talk about that during my Phd. thesis. 

Since about 3 years now, I tried to write a good paper about this subject and the fundaments of my research prototype Btrplace. After tons of rejections, and even more tons of improvements, my paper finally passes the wall. It has been accepted to a special issue of the IEEE Transactions on Dependable and Secure Computing related to Cloud Computing assessment. The manuscript should appear in April 2013. The preprint is however already available here.

Last, I am currently at the national conference COMPAS. Yesterday, I talked about Btrplace at the workshop on SLA Management in Cloud Computing. The slides are available here.

The Btrplace Sandbox

The VM placement algorithm inside Entropy is no longer a child and it deserves its own name now: Btrplace.

To celebrate this announcement, I have developed a sandbox that allows users to play with the algorithm on a simulated datacenter directly from the browser. It's simple, it's fun and it is available here.

This sandbox has been realized in the context of the FUI openCloudware project. This project aims at building an open software engineering platform for the collaborative development and deployment of distributed applications.

Back from TridentCom 2012

Hi

I just went back from the 2012 edition of TridentCom.

The conference was great with very interesting papers and a lot of demos. The swimming pool was nice too :D.

Icing on the cake, the paper I made with Robert Ricci get the best paper award. On my publications page, you can now have an access to the author copy of the paper and the slides I presented during the conference.

Entropy + Fit4Green = ♥

Fit4Green is a project that is funded by the Commission of European Union. It's main goal is to reduce the direct energy consumption of ICT resources of a federation of data centres.

Since mid 2011, the resource optimizer of the Fit4Green prototype relies on the plan module of Entropy, called BtrPlace, to compute the VM placement and the server state according to various green concerns.

A research paper describing their work has been accepted to the E-energy conference. The author version of the paper is available on my publication page, or directly here.

CSI - Salt Lake City: How To Build a Better Testbed

During my postdoc at Flux. I have analysed the usage of Emulab since its creation. With Robert Ricci, we used these informations to try to improve the testbed design with regards to its real usage rather than assumptions. The result of this investigation has been accepted to the TridentCom 2012 conference.

If you always wanted to see a crime scene, the paper is available from my website or directly here.

Beware sherlock holmes, here we are !

Bin Repacking Scheduling in Virtualized Datacenters: Back to work

As promised, our recent modifications in the code and in the model of the VRSP are available online. With regards to the results published in CP 2011, we improved the solving capabilities of Entropy while providing better solutions. Icing on the cake, solutions are computed about 10 times faster !

The report is available from my website or directly here.