GreenMap: MapReduce with ultra-high-efficiency power delivery

With the continuous growth of online services, energy consumption has become a significant fraction of the total cost of ownership of large data centers. Though much work in green computing has focused on improving efficiency for computation units such as CPU’s or servers, little attention has been paid to power delivery structures, such as voltage converters, which takes 10-20% of total energy consumption even before any computation takes place. Recently, a new power delivery architecture called series stack has been proposed in the power community, aiming to reduce conversion power loss. In series stack, servers are connected serially, and differential converters are used to regulate server voltage. However, to effectively reduce conversion loss in series stack, computation loads need to be balanced in real time. To balance load for series stack, we implemented GreenMap, a modified MapReduce framework on top of series stacks, that assigns tasks in synchronization. We evaluated the conversion loss of GreenMap on a small data center. At all loads, GreenMap achieves a 81x-138x reduction in conversion loss from commercial-grade high voltage converters used by today’s data centers. The saved power is equivalent to 15% reduction in total energy consumption. GreenMap also achieves 67%-80% reduction in conversion loss compared to Hadoop’s FIFO scheduler under serial stack structure. Based on the observation that the average response time of GreenMap suffers a degradation at low load, we further propose a modification of GreenMap with dynamic scaling to achieve a favorable tradeoff between response time and power efficiency.

Name thesis:degree_name

M.S.

Level thesis:degree_level

Thesis

Discipline thesis:degree_discipline

Electrical & Computer Engr

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2018

Author dc:creator

• Su, Du

Contributors dc:contributor

• Lu, Yi

• Green computing, MapReduce, Load Balance

Statement dc:rights

• Copyright 2017 Du Su

Language dc:language

en

Handle dc:identifier

http://hdl.handle.net/2142/100882

OAI identifier oai:identifier

oai:www.ideals.illinois.edu:2142/100882