It's well-known that data centers are extremely power hungry. In the U.S., operators need to spend roughly 10 billion USD or more on their electricity bills each year. Admittedly,
a handful of top-brand IT firms, e.g., Google, Facebook, Microsoft, and Apple, are doing great jobs in data center energy efficiency. But, they
seem to have delivered to the public a “misleading” signal: data centers are already energy-efficient enough.
The existing efforts led by those Google-type IT giants as well as the research community have been primarily focused on those
“hyper-scale data centers” (e.g., Google's data centers), which, however, are just a very small segment of the data center industry.
I compiled the estimated percentage of electricity consumption by the U.S. data center segment in 2011 (shown above), excluding those
small server closets and rooms, which claim themselves as “data centers” and actually take up nearly 50% of all the data center energy consumption.
My research focuses on power-management in multi-tenant colocation data centers
and questions the wide perception “data centers are already energy-efficient” (delivered by Goolge-type IT giants). A few examples of colocation
data centers are Equinix, Digital Realty, Verizon Terremark, CoreSite, etc. These data centers are “hidden” from the public and hence rarely investigated by the
research community, but they constitute a much bigger segment of the data center industry than those so-called hyper-scale cloud data centers
that have received a predominant amount of research attention. Further, compared to Google-type data centers,
colocations, as well as a vast majority of enterprise/corporate data centers,
are falling far behind in energy efficiency.
More interestingly, multi-tenant colocation data
centers present unique challenges that cannot be addressed by the existing solutions! For more details, please refer to the project page:
Paper “GreenColo: A Novel Incentive
Mechanism for Minimizing Carbon Footprint in Colocation Data Center” will appear in International Green Computing Conference (IGCC) 2014!
Paper “A First Look at Colocation Demand Response”
will appear in Greenmetrics 2014 (extended summary of ICAC’14 paper)! [PDF]
Paper ‘‘A Theoretical Foundation for Scheduling and Designing Heterogeneous Processors for Interactive Applications’’ will appear in
28th International Symposium on Distributed Computing (DISC) 2014! [Longer Version]
Journal paper ‘‘Dynamic Scheduling for Energy Minimization in
Delay-Sensitive Stream Mining’’ will appear in
IEEE Transactions on Signal Processing.
Two papers will appear in USENIX International Conference on Autonomic Computing (ICAC) 2014!
Article “How Can Supercomputers Survive a Drought?” appears at
GreenBiz covered my research: “5 reasons the thirst for water technology will grow in 2014”
Power Management in Colocation Data Centers [PDF]
Power-hungry data centers have been massively expanding in both number and scale, placing an
increasing emphasis on optimizing data center power management. While the progress in data
center energy efficiency is encouraging, the existing efforts have dominantly centered around
owner-operated data centers (e.g., Google). Another unique and integral segment of data center
industry — - colocation data center, simply called “colocation”, which is the physical home to
many Internet and cloud services – has not been well investigated, which, if still left unchecked,
would become a major hurdle for sustainable growth of the digital economy.
In sharp contrast
with owner-operated data centers where operators have full control over both computing
resources and facilities, colocation rents physical space to multiple tenants which individually
control their own physical servers and power management, while the colocation operator is mainly
responsible for facility support (e.g., providing reliable power and cooling). The uncoordinated
power management, resulting from the colocation operator's lack of control over tenants’ servers,
invalidates many of the existing power management solutions for owner-operated data centers,
thereby making colocations’ operation highly inefficient.
Making Data Centers Less ‘‘Thirsty''
Performance-driven budget-constrained autoscaling
Scheduling interactive services on heterogeneous processors
Shaolei Ren received the B.E. degree in Electronic Engineering from Tsinghua University in
July 2006, the M.Phil. degree in Eletronic and Computer Engineering from Hong Kong
University of Science and Technology in August 2008, and the Ph.D. degree in
Electrical Engineering from University of California, Los Angeles, in June 2012.
Since August 2012, he has been with Florida International University,
where he currently holds a joint appointment of Assistant Professor in the School of Computing and Information Sciences and the Department of Electrical and Computer Engineering.
Dr. Ren directs the Sustainable Computing Group at FIU. His research centers around cloud computing and data center resource management,
with an emphasis on sustainability. He received the Best Paper Award at
the 8th International Workshop on Feedback Computing, the Best Paper Award at
IEEE International Conference on Communications in 2009, and was selected by
IBM T. J. Watson Research as one of the 10 Emerging Leaders in Multimedia and Signal Processing in 2010.
School of Computing and Information Sciences
Florida International University
11200 SW 8th Street, ECS 350
Miami, FL 33199