Loading presentation...

Present Remotely

Send the link below via email or IM

Copy

Present to your audience

Start remote presentation

  • Invited audience members will follow you as you navigate and present
  • People invited to a presentation do not need a Prezi account
  • This link expires 10 minutes after you close the presentation
  • A maximum of 30 users can follow your presentation
  • Learn more about this feature in our knowledge base article

Do you really want to delete this prezi?

Neither you, nor the coeditors you shared it with will be able to recover it again.

DeleteCancel

Make your likes visible on Facebook?

Connect your Facebook account to Prezi and let your likes appear on your timeline.
You can change this under Settings & Account at any time.

No, thanks

Green Computing in Data Centers

No description
by

Shen Li

on 20 April 2015

Comments (0)

Please log in to add your comment.

Report abuse

Transcript of Green Computing in Data Centers

Turning off servers when workload is light: A.K.A. Dynamic Provisioning
Green Computing in Data Centers
Methodologies
Turning off servers when the workload is light

Scaling down CPU frequencies when it does not need that much computation resources

Shifting workload around different data centers to take advantages of lower electricity price

Using batteries to reduce peak power consumption

Postpone delay-insensitive workload till when the data centers power demand is low.

Rely on green energy

Build more "fitted" and power-efficient servers

Assign more workload to servers that are easy to cool

Use out door air to chill servers when possible

and many more....
Rely on Green Energy
Reduce peak power consumption
A.K.A.: Power Capping
Background
Reduce as much energy cost as possible without affecting FB user experiences.
Turning off servers when the workload is light

Scaling down CPU frequencies when it does not need that much computation resources

Shifting workload around different data centers to take advantages of lower electricity price

Using batteries to reduce peak power consumption

Postpone delay-insensitive workload till when the data centers power demand is low.

Rely on green energy

Build more "fitted" and power-efficient servers

Assign more workload to servers that are easy to cool

Use out door air to chill servers when possible

and many more....
1: powered on
0: powered off
f: cpu frequency
p = 3
power consumption of server i
idle power consumption
constant parameter
number of servers
total power consumption of the cluster
minimize
subject to
total workload
Server i's workload
Desired response time
response time function taking service rate and arrival rate as parameter
Constant parameter
Integer Programming is NP-hard
order all servers according to efficiency [ ]. Servers are turned on following this order.
Divide workload range into
such that the first k servers needs to be on
when fall in range
To get rid of
L. Wang, Y. Lu, Efficient Power Management of Heterogeneous Soft Real-Time Clusters, RTSS 2006
CPU intensive
Memory intensive
Disk intensive
Is there any problem here?
What is the penalty of turning off machines?
What about cache servers or distributed file systems?
S. Li, S. Wang, F. Yang, S. Hu, F. Saremi, T. Abdelzaher, Proteus: Power Proportional Memory Cache Cluster in Data Centers, ICDCS 2013
http://memcached.org/
R. Kaushik, M. Bhandarkar, GreenHDFS: Towards an Energy-Conserving Storage-Efficient, Hybrid Hadoop Compute Cluster. HotPower 2010.
Cannot turn off more than 2 DataNodes.
What is power capping?
Why batteries?
Facebook
Wikipedia
Akamai
Workload Pattern
D. Wang, C. Ren, A. Sivasubramaniam,B. Urgaonkar, and H. Fathy, "Energy Storage in Datacenters: What, Where, and How much?", SIGMETRICS 2012.
D. Palasamudram, R. Sitaraman, B. Urgaonkar, R. Urgaonkar, "Using Batteries to Reduce the Power Costs of Internet-scale Distributed Networks", SoCC 2012.
minimize
subject to
Peak power cost
Battery cost
Power demand has to be satisfied
Power does not flow from server to grid or battery.
Power cap cannot be violated
Cannot violate battery charging rate
How battery state changes
Cannot violate battery capacity
Inigo Goiri, Kien Le, Thu Duc Nguyen, Jordi Guitart, Jordi Torres, Ricardo Bianchini, "GreenHadoop: Leveraging Green Energy in Data-Processing Frameworks", EuroSys 2012.
What is Map-Reduce?
J. Dean and S. Ghemawat, "MapReduce: Simplifed Data Processing on Large Clusters", OSDI 2004.
Thank you!
Shen Li
shenli3@illinois.edu
D. Wang, C. Ren, A. Sivasubramaniam,B. Urgaonkar, and H. Fathy, "Energy Storage in Datacenters: What, Where, and How much?", SIGMETRICS 2012.
Year
Power (GWatt)
Global Data Center Power Demand
source: DCD Intelligence 2012
source: Lawrence Berkeley National Laboratory
We need to consider both cyber issues (like load distribution/scheduling) and physical issues (like cooling).
How to solve this CPS problem? What's your call?
Recap
9 methodologies to reduce energy cost in data centers
Dynamic provisioning for web servers, cache servers, and HDFS respectively .
Typical data center power infrastructure, and how batteries help to reduce the power cap.
MapReduce framework and GreenHaoop
Q: What facts motivate each methodology?
Q: Why do we need different solutions for different systems?
Q: What will happen if the grid power fails?
Q: How does MapReduce work? How does GreenHadoop help to cut brown energy cost?
Electricity Bill
Full transcript