Google presents paper on how to cut electricity in the data centre
Share on Twitter.
Get the most reliable SMTP service for your business. You wished you got it sooner!
June 6, 2014
Google has developed a method to save as much as 20 percent of the electricity used to power
its multiple data centers by reaching deep into the core of its infrastructure and experimenting
with different formulas.
In a paper to be presented next week at the ISCA 2014 computer architecture conference entitled "Towards
Energy Proportionality for Large-Scale Latency-Critical Workloads", researchers and engineers
from Google and Stanford University discuss an experimental system named "Pegasus" that may save
Google large amounts of cash by helping it cut its electricity consumption.
To be sure, Pegasus addresses one of the worst-kept secrets about cloud computing, which is
that the computer chips in the gigantic data centers of Google, Amazon and Microsoft are standing
idle for significant amounts of time.
Though all these companies have developed sophisticated technologies to try to increase the
utilization of their chips, they all fall short in one way or another.
This simply means that a substantial amount of the electricity going into their data centers
is completely wasted as it powers server chips that are either completely idle or in a state of very
From an operator's perspective, it's a bit of a losing proposition, and from an environmentalist's
perspective it's a real blunder.
Now Google and Stanford researchers have designed a new system that increases the efficiency of the
power consumption of the data centers, but without compromising performance in any way.
Pegasus does this by dialing up and down the power consumption of the processors within Google's
servers according to the desired request-latency requirements – dubbed iso-latency – of any
The power management technology that Pegasus uses is 'Running Average Power Limit' or RAPL, which
allows you to incrementally tweak CPU power consumption in amounts of just 0.125W.
The system "sweeps the RAPL power limit at a given load to find the point of minimum cluster
power that satisfies the service-level objective target".
Put another way, Pegasus makes sure that a processor is working just hard enough to meet the
demands of the application running on it, but nothing else.
"The baseline can be compared to driving a car with sudden stops and starts. Having said that, iso-latency
would then be driving the car at a slower speed to avoid accelerating hard and braking harder," the
"The second method of operating a car is much more fuel ef?cient than the ?rst, which is akin to
the results we have observed," the paper suggests.
To be sure, existing power management techniques for large data centers advocate turning off
individual servers or even individual cores, but the researchers said this was inefficient, and it is.
"Even if spare memory storage is available, moving tens of gigabytes of state in and out of
servers is expensive and very time consuming, making it dificult (if not impossible) to react
to fast or small changes in load," they explain.
Shutting down individual computer cores, meanwhile, doesn't work due to the specific needs
of Google's search technology. "A single user query to the front-end will lead to forwarding of the
query to all leaf nodes. As a result, even a small request rate can create a non-trivial amount of
work on each server," say the researchers.
"For instance, consider a cluster sized to handle a peak load of 10,000 queries per second (QPS)
using 1000 servers," they explain. "Even at a 10 percent load, each of the 1000 nodes are seeing on
average one query per millisecond. There is simply not enough idleness to invoke some of the more
effective low power modes," they explain.
So, PEGASUS, which stands for Power and Energy Gains Automatically Saved from Underutilized Systems,
has been created. The technology is a dynamic, feedback-based controller that enforces the iso-latency
It tweaks the power to the chip according to the task it's running, making sure to not violate
any service-level agreements on latency.
During various tests on Google's production workloads, the researchers found that PEGASUS saved
as much as 30 percent of power compared to a non-PEGASUS system during times of low demand, and
11 percent total energy savings over a 24-hour period.
The team also evaluated it on a full scale, production cluster for its search engine at Google.
There, Pegasus did marginally less by saving between 10 and 20 percent on average during low
This is due to the way it applied policy across the thousands of servers without taking into
account variations between chips.
A potential solution to this is to distribute the PEGASUS controller so that it lives on each
node and applies latency policy from there.
"The solution to the so-called 'hot leaf issue' is fairly straightforward-- implement a distributed
controller on each server that keeps the leaf latency at a certain latency goal," the researchers write."
In the real world, as any grey haired veteran of distributed systems can tell you, implementing
any kind of distributed controller scheme is akin to inviting a world of confusion and pain into your
However, Google is a gold-plated organization that can fund the necessary engineers to keep a
distributed scheme like this working. And it does work.
If PEGASUS were to be implemented in a distributed way, the researchers say it could save up to
35 percent of power over the baseline-– a huge savings for a company the size of Google.
As is typical with Google, the paper gives no details of whether PEGASUS has been deployed across
Google's infrastructure in production, but given these huge power savings and the substantial amount
of work Google has invested in the technology, we think it's very likely.
"Overall, iso-latency provides a signi?cant step forward towards the goal of energy proportionality
for one of the challenging classes of large-scale, low-latency workloads," the researchers write.
The deployment of complex systems like PEGASUS alongside other advanced Google technologies such
as OMEGA (cluster management), SPANNER (distributed DBMS), or CPI2 (thread-level performance monitoring)
enables Google to make its data centers dramatically more efficient than those operated by smaller, less
These technologies will, over time, help Google compete in public cloud with rivals such as Amazon
and Microsoft, while serving more ads at a lower cost than before.
In other IT news
For the past few months, better code obfuscation has attracted the attention of the prestigious
Association of Computing Machinery, which has singled out a developer working at IBM's T.J. Watson Research Centre
with an award for his work.
On any given day, protecting programming code, even as a binary, from being reverse-engineered
is very difficult-- any method that encrypts the code has to keep its functionality in place, and
decrypting the code for execution has to be extremely fast in order for the protection to work seamlessly.
Sanjam Garg, an alumni of the Institute of Technology of New Delhi, India, claims to have cracked
that issue in his recently published paper, Candidate Multilinear Maps from Ideal Lattices.
As that paper explains, bilinear maps are very well-known, and their applications are “too numerous”
to list but tripartate Diffie-Hellman and identity-based encryption are two examples of how it's done.
Expanding that concept to multi-lineal maps has been theorized, and Garg now writes it in his paper, but it wasn't
previously achieved, however.
That work was then expanded in collaboration between Garg and researchers from Microsoft, Boston University
and UCLA which demonstrated that Garg's concepts are workable for program code obfuscation.
As they put it in the paper's abstract, Garg's work provides a “candidate obfuscator that
cannot be broken by algebraic attacks”.
As the ACM notes-- “Garg described new mathematical tools that serve as key ingredients for transforming
a program into a jigsaw puzzle of encrypted pieces.
Corresponding to each and every single input is a unique set of puzzle pieces that, when assembled,
reveal the output of the program.
Get the most dependable SMTP server for your company. You will congratulate yourself!
Share on Twitter.