Personal tools

Efficient resource allocation in data centers

Motivation:

Resource allocation is a crucial aspect in both small and large data centers. Ideally, the level of resources consumed should be proportional to the system load at any time, rather than to the peak load the data center is prepared for, and resources should be allocated to optimize for load balancing and power consumption. This project investigates resource allocation techniques which are capable of optimally processing application requests and exploiting variations in their load (e.g., switch off parts of the data center when the load is low, or add more resources when the load is high).

Goals:

Get familiar with the Rhizoma's resource monitoring infrastructures and its CLP (constraint logic programming) solver. Design and implement techniques for resource allocation which satisfy the request requirements while optimizing metrics such as load balancing and data center's power level. Explore how CLP can be used to specify the resource allocation problem, and to provide a more expressive request representation (both for resource request and request advertisement) in data centers. Possibly extend the Rhizoma's sensor infrastructure (currently reporting latency, cpu, memory information) and integrate the implemented techniques in the Rhizoma overlay. Experimentally evaluate the work using PlanetLab and in-house clusters.