A PDF file should load here. If you do not see its contents
the file may be temporarily unavailable at the journal website
or you do not have a PDF plug-in installed and enabled in your browser.
Alternatively, you can download the file locally and open with any standalone PDF reader:
http://link.springer.com/content/pdf/10.1007%2Fs11227-010-0485-0.pdf
Energy-efficient high-performance parallel and distributed computing
Samee Ullah Khan
0
Pascal Bouvry
0
Thomas Engel
0
0
S.U. Khan NDSU-CIIT Green Computing and Communications Laboratory, Electrical and Computer Engineering Department, North Dakota State University
, Fargo,
ND 58108-6050, USA
-
High-Performance Computing (HPC) is a major contributor to cutting-edge research
and discovery in science and technology. We can attribute several key research
findings that were aided or validated by tests and simulations run on HPCs. Over the last
decade, we have witnessed computing service providers to continually upgrade their
infrastructures to HPCs that can meet the increasing demands of powerful newer
applications. In parallel, almost in concert, computing manufacturers have consolidated
and moved from stand-alone servers to rack mounted blades. The aforementioned
trends alone are increasing electricity usage in large-scale computing systems, such
as data centers, computational grids, and cloud computing. This increase in electricity
utilization has reached to a point that many information technology managers are all
up in arms to identify a holistic solution that can reduce electricity consumption (so
that the total cost of operation is minimized) of their respective large-scale computing
systems and simultaneously improve upon or maintain the current throughput of the
system.
2 Research problems in energy-efficient HPCs
We must understand that in HPCs the energy efficiency must not come at the expense
of performance. Therefore, innovative methodologies must be conceived that can
balance energy consumption and performance. The key research issue remains in
providing a holist solution that can collectively minimize energy consumption of a HPC
facility. Researchers have fundamentally focused in isolation on improving energy
consumption of the: (a) optimizing the room cooling system, (b) processing elements
by scheduling and mapping, (c) network by intelligent routing, and (d) memory by
reducing data migration. Because performance and energy consumption must be
balanced in a HPC environment, solutions that salvage energy at all of the three
computing components must be considered. This is a difficult task because the complexity of
the meta model to capture such interactions will increase astronomically. This also is
the reason why we consider such an approach a key research issue in Energy-efficient
HPCs.
3 Selected papers for the Special Issue
Several high quality research articles were submitted. We (the guest editors) had a
very difficult decision to make on the inclusion and exclusion of research articles.
Because of the space constraints, we could only include the following five outstanding
research articles.
Proactive Thermal Management in Green Datacenters proposes an intelligent
temperature regulating mechanism for data centers.
Improving Performance and Energy Efficiency of Embedded Processors via
PostFabrication Instruction Set Customization introduces an adaptive extensible
processor in which custom instructions are generated and added after chip-fabrication to
reduce energy consumption and improve performance.
Energy Efficient Scheduling of Parallel Tasks on Multiprocessor Computers
proposes a methodology to schedule parallel tasks on multiprocessors using dynamic
voltage and speed scaling.
Reliability-Aware Platform Optimization for 3D Chip Multiprocessors proposes
an optimization methodology that integrates power, performance, and temperature for
multiprocessor systems.
Energy Efficient Utilization of Resources in Cloud Computing Systems
introduces a task consolidating technique to mitigate energy consumption in under utilized
large-scale computing systems.
(...truncated)