Locality-aware Cooperation for VM Scheduling in Distributed Clouds
Résumé
The promotion of distributed cloud computing infrastructures as the next platform to deliver the Utility Computing paradigm, leads to new virtual machines (VMs) scheduling algorithms leveraging peer to peer approaches. Although these proposals considerably improve the scalability, leading to the management of hundreds of thousands of VM over thousands of physical machines (PMs), they do not consider the network overhead introduced by multi-site infrastructures. This overhead can have a dramatic impact on performance if there is no mechanism for favoring intra-site vs. inter-site manipulations. This paper introduces a new building block designed over a Vivaldi over- lay which maximizes efficient collaborations between PMs. We combined this mechanism with DVMS, a large scale virtual machine scheduler and showed its benefit by discussing several experiments performed on four distinct sites of the Grid'5000 testbed. Thanks to our proposal and with- out changing the scheduling decision algorithm, the number of inter-site operations has been reduced by 72%. This result provides a glimpse of the promising future of locality properties to improve performance of massive distributed cloud platforms.