VM Placement strategies for cloud scenarios
Nicol˛ Maria Calcavecchia
DEI PhD Student
DEI - Sala Seminari
10 Novembre 2011
The increasing diffusion of cloud computing paradigm is fostering the creation of massive data-centers composed by hundreds of thousands of computing elements. In this context, virtualization technology is typically used to effectively reduce potential wastes of resources due to over-provisioning by consolidating more virtual machines (VMs) within a single physical server. Virtual machine placement addresses this problem by minimizing the number of powered-on hosts while maintaining a high level of performance. However, the majority of the existing works ignore the dynamic nature of the incoming stream of VM deploy requests under which the cloud infrastructure is subject over time.
In this seminar we present a novel technique for VM placement that, by monitoring past behaviour of VMs, is able to efficiently produce a new placement with a reduced number of migrations. Differently from other work in literature, we explicitly evaluate our technique against a typical cloud scenario with a continuous stream of VM deployments. Our results are compared with placements produced by a Mixed Integer Linear Programming model solved through the CPLEX optimization tool.
Area di ricerca:
Metodologie e architetture software avanzate