Authors:
Shengquan Wang
1
;
Waqaas Munawar
2
;
Xue Liu
3
and
Jian-Jia Chen
2
Affiliations:
1
Univ. of Michigan-Dearborn, United States
;
2
Karlsruhe Institute of Technology, Germany
;
3
McGill University and Canada, Canada
Keyword(s):
Power Saving, Server Farm, Multi-tier, Response Time, Service Level Agreement (SLA), M/G/1/PS, Mean Value Analysis (MVA).
Related
Ontology
Subjects/Areas/Topics:
Energy and Economy
;
Energy-Aware Systems and Technologies
;
Optimization Techniques for Efficient Energy Consumption
Abstract:
Server farms suffer from an increasing power consumption nowadays.
Power saving has become a prominent design issue in server farms.
This paper presents a power-saving design in server farms
under the constraint of the response time.
In particular, we target on multi-tier applications, which are
very typical on the web in modern days.
We propose an efficient power-saving design strategy, called PowerTier.
This strategy exploits two major techniques by using Dynamic Power management (DPM)
to activate/deactivate servers and using Dynamic Voltage Scaling (DVS) to adjust
the processor speed for each activated server.
In addition, PowerTier considers two different application models:
the open-queueing model
and the closed-queueing model for session-less and session-based web
applications respectively.
With PowerTier, we are able to choose the number
of activated servers at each tier and the processor
speed for each server to minimize the overall power consumption
in server farms while m
eeting a given mean response time
guarantee for multi-tier applications.
Our comprehensive simulation confirms the effectiveness and efficiency
of PowerTier.
(More)