Power saving in data centers

Power saving in data centers
The future of TIC
Álvaro López Medina
ETSIT – UPM 2015

1 Introduction
PUE2
Cognitive Networks3
Conclusions4

INTRODUCTION
What is a data center?
A data center is a facility that
stores and manages lots of
information of one or more
companies.
Some companies, like Google
and Amazon, rent some
servers from its own data
centers to other companies.
This fact concentrates all the
energy consumption in
several big data centers.
1 Introduction

INTRODUCTION1
Google, Facebook, Microsoft and Amazon are the companies
with more data center facilities around the world.
Google has 900 000
servers distributed in
13 data centers.
Amazon has 450 000
servers, 40 000
dedicated to their Web
Services division.
Microsoft, with over
500 000 servers, has
spent over $15 billion
on its data centers.
Facebook stores 100
Petabytes of data,
around 100 000 000
GB, in their data
centers.
Relevant data
1 Introduction

INTRODUCTION1
This large number of servers and information stored in the data
centers demands a large amount of energy.
Google uses around 260 million watts which accounts to
0.01% of global energy. This power is enough to supply
200 000 homes.
The second largest data center (Quality Technology,
Atlanta) is supported by 80 megawatts of power capacity.
This is a problem that TIC may solve in the future.
The standard storage rack (*) in Facebook uses 8 Kilowatts
of power.
The power density of a data center is 100 times more
than that of a large commercial office building.
(*) A rack is a enclosure for mounting multiple equipment modules (around 10 servers).
4
1
2
3
Energetic consumption
1 Introduction

PUE2
These numbers make us to consider whether power consumption of the data
centers is used efficiently or a significant part is wasted. To classify data
centers concerning their energy efficiency it is used the Power Usage
Effectiveness (PUE).
- The Total Facility Energy includes: IT Equipment Energy, lighting, cooling…
- IT Equipment Energy includes only the energy consumption of the servers.
The average data center PUE is about 1.8 which means that the data center
has an over cost of 80% in activities that are not exclusive of the servers
(mainly due to cooling activities).
PUE =
Total Facility Energy
IT Equipment Energy
Energy Efficiency
2 PUE

PUE2
Now, companies have realized that whereas having 10 inefficient
servers may not be an issue, when owning 10 000 or even 100 000
servers their efficiency means saving much more costs.
“The aim of innovation in the future will be focused on
minimizing the energetic consumption in Data Centers”
Scope
2 PUE

PUE2
There are three main ways to improve energy saving:
Using less consumption components. There are lots of
innovations but it is just keep working with the same idea.
Changing the layout of the servers so that there are cool
currents and hot currents that don’t mix each other.
Using cognitive networks to make all the servers think as
one.
1
2
3
Ways of improve the PUE
2 PUE

PUE2
Changing the layout of the servers to create corridors between them.
These corridors will be of two types:
 The cold ones will take the fresh air from the air conditioning
system to the entry of each server.
 The hot ones collects the hot air that leaves the servers to lead
it to the air conditioning system again.
Changing the layout
2 PUE

PUE2
There are three main ways to improve energy saving:
Using less consumption components. There are lots of innovations
but it is just keep working with the same idea.
Changing the layout of the servers so that there are cool currents
and hot currents that don’t mix each other.
Using cognitive networks to make all the servers think as one.
1
2
3
This is the most important way
Ways of improve the PUE
2 PUE

COGNITIVE NETWORKS3
“A cognitive network is a
network formed by a large
amount of sensors and
actuators that make a great
Artificial Intelligence (AI)
which monitor the
consumption and the needs of
the servers and modify the set
up so that they save as much
energy as possible”
What is it?
3 Cognitive Networks

COGNITIVE NETWORKS3
Sensors monitor constantly the
environment: Humidity, temperature,
load of the servers, r.p.m. of the server
fans, consumption of the power source,
etc.
Then, they send it to a central server
which collects all the information.
The main server, which is an AI, process
the data to know if the system is working
properly or is underusing or oversizing
the resources it has.
The central server communicates to all
the servers and actuators, which energy
consumption policies they should follow
to maximize the capacity of the servers
and minimize the power consumption.
4
1
2
3
How it works?

COGNITIVE NETWORKS3
It will centralize the monitoring of all the information of
the data center.
It will have a centralize governor policy that will manage
all the servers.
It will have real-time information of every component of
the servers.
All the servers will work as a unique entity which has
perfect coordination and able to take important
consumption decisions.
4
1
2
3
Thanks to Cognitive Networks, the data center that implements
it will achieve the following advantages:
What advantages does it offer?

CONCLUSIONS4
During this presentation we have seen how important data
centers are in all the businesses, how they impact in the
power consumption of the world and the need to improve
the energy saving.
So, as a conclusion of what we have discussed all along this
presentation, we can safely say that THE FUTURE OF TIC will clearly
involve making an extraordinary effort in order to implement the
technical solutions mentioned above.
4 Conclusion

Power saving in data centers

In this document