COL:Anshul Gandhi

Event Type: 
Wednesday, March 13, 2013 - 14:30

Gandhi, Anshul Gets: March 13 (Wed)


Dynamic Capacity Management for Multi-Tier Data Centers


Online services, such as Amazon and Facebook, are provided by multi-tier data center infrastructures. A primary goal for these service providers is to meet certain response time Service Level Agreements (SLAs). In an effort to meet these SLAs, service providers typically over-provision the number of servers to meet their estimate of peak demand, and leave these servers always on. Such static approaches clearly waste a lot of power, and further, can be quite expensive if the servers are rented from a cloud computing provider on a pay-per-use basis.

In this talk, I will present our capacity management policy, AutoScale, that greatly reduces the number of servers needed by applications driven by unpredictable, time-varying demand, while meeting response time SLAs. AutoScale matches incoming demand by dynamically scaling capacity up or down, as needed.

We evaluate AutoScale via implementation on a 38-server multi-tier data center, serving a web site of the type seen in Amazon or Facebook, with a key-value store workload. We find that AutoScale improves upon the current static capacity management policy used in data centers by up to 50% with respect to power and resources, and furthermore, vastly improves upon existing dynamic capacity management policies with respect to meeting SLAs and robustness.

Short bio:

Anshul Gandhi is a Ph.D. student in the Computer Science Department at Carnegie Mellon University, under the direction of Mor Harchol-Balter. His research involves designing and implementing power management policies for datacenters as well as general performance modeling of computer systems.

Computed Event Type: