Profit-Aware Server Allocation for Green Internet Services

In recent years large investments have been made to build data centers, or server farms, purpose-built facilities providing storage and computing services within and across organizational boundaries. A typical server farm may contain thousands of servers, which require large amounts of power to operate and to keep cool, not to mention the hidden costs associated with data centers’ carbon footprint and water consumption for cooling purposes. On the other hand, the increasing use of the Internet as a provider of services and a major information media have changed significantly, especially over the last ten years. Expectations in terms of performance and responsiveness have markedly grown. Consequently, it is very important to devise strategies aiming at reducing the power consumption while maintaining acceptable levels of performance.
In “Profit-Aware Server Allocation for Green Internet Services” we examine the market in computer services by introducing a quantitative framework where the performance of a provisioning system with setup cost and serving impatient customers is measured by the average revenue earned per unit of time. Every completed request generates a certain amount of profit, running servers consume electricity for power and cooling, while waiting customers might leave the system before receiving service if they experience excessive delays. A dynamic allocation policy aiming at satisfying the conflicting goals of maximizing the quality of users’ experience while minimizing the cost for the provider is introduced and evaluated. Since the revenue of the system is affected by its energy consumption, hardware maintenance costs and the amount of successfully served customers, the policy we propose aims at minimizing the first two factors, while maximizing the last one. The results of several experiments are described, showing that the proposed scheme performs well under different traffic conditions. In particular, we find that the number of running servers has a significant effect on the revenue earned by the provider (this is very important, given that the average utilization is in the 10 to 20% range, while electricity prices keep rising).
In “On Allocation Policies for Power and Performance” we analyze the Wikipedia traces, and present various heuristics capable of dealing with non-stationary user demand which can not be predicted with absolute accuracy. The experiments we have performed show that the policies we have developed are not very sensitive to errors in parameters’ estimation (i.e., they perform well even in spite of sudden changes of the user demand).

Advertisements

About michelemazzucco

Michele Mazzucco is a Post-Doctoral Research Fellow in Software Engineering at the University of Tartu (Estonia). He conducts research in distributed systems, performance engineering and green computing.

Trackbacks/Pingbacks

  1. Research group digest – ulno.net - March 18, 2011

    […] Profit-Aware Server Allocation for Green Internet Services In recent years large investments have been made to build data centers, or server farms, purpose-built facilities providing storage and computing services within and across organizational boundaries. A typical server farm may contain thousands of servers, which require large amounts of power to operate and to keep cool, not to mention the hidden costs associated […] […]

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

%d bloggers like this: