|Ph.D Student||Funaro Liran|
|Subject||Market Driven Multi-Resource Allocation|
|Department||Department of Computer Science||Supervisors||Professor Assaf Schuster|
|Dr. Orna Agmon Ben-Yehud|
|Full Thesis text|
Suboptimal resource utilization among public and private cloud providers prevents them from maximizing their economic potential.
Long-term allocated resources are often idle when they might have been subleased for a short period.
Providers could address this problem by overcommitment of their resources, but this may lead to unpredictable client performance if all clients try to use the resources simultaneously.
A better alternative would be for cloud providers to allocate their physical resources to their clients dynamically, as needed, thereby maximizing the benefit that the former get out of given hardware, and maintaining the latter’s satisfaction.
In this work we focus on economic resource allocation mechanisms for maximizing the provider's profits and the aggregated value all clients draw from the cloud.
We provide two approaches to achieving this objective: an auction-based mechanism, and a stochastic resource allocation coupled with a smart pricing scheme.
In developing these mechanisms, we had to overcome a number of challenges.
One challenge involved the high computational complexity of the optimization problem required by the auction mechanism.
To solve it, we developed an efficient auction algorithm.
The other challenge was the lack of benchmarks to evaluate a memory allocation scheme.
We needed applications whose performance is proportional to the memory availability, i.e., memory elastic applications.
Accordingly, we compiled a set of memory elastic benchmarks, as well as a language and a method to evaluate their elasticity.