6/4/2018 1 THE DATACENTER AS A COMPUTER George Porter Week 10 Spring 2018 ATTRIBUTION • These slides are released under an Attribution-NonCommercial-ShareAlike 3.0 Unported (CC BY-NC-SA 3.0) Creative Commons license • These slides incorporate material from: • The Datacenter as a Computer: An Introduction to the Design of Warehouse-Scale Machines, 2nd ed., by Barroso, Clidaras, and Hölzle.
Welcome message from author
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
CSE 124: DatacentersGeorge Porter Week 10 Spring 2018
ATTRIBUTION • These slides are released under an Attribution-NonCommercial-ShareAlike 3.0
Unported (CC BY-NC-SA 3.0) Creative Commons license
• These slides incorporate material from:
• The Datacenter as a Computer: An Introduction to the Design of Warehouse-Scale Machines, 2nd ed., by Barroso, Clidaras, and Hölzle.
6/4/2018
2
ANNOUNCEMENTS (1/2) Please read The Datacenter as a Computer: An Introduction to the Design of Warehouse-Scale Machines, 2nd ed., by Barroso, Clidaras, and Hölzle.
Ideally all of it, but at least Chapters 1-5
Sample finals posted to Piazza’s “resources” page
“Minimal” Homework 5 posted
PROJECT 2 ANNOUNCEMENTS (2/2)
Hash function: use provided function (pass bytes into it instead of string)
Distributed getversion test removed
Download text file, open with text editor to make sure it is ok
Client to/from MDS: we will use our own client (make sure to follow RPC given, don’t create your own)
Will manually look for 2PC implementation (60% not implementing 2PC…)
6/4/2018
3
• Pages 221-224
• Data flows along tree, not underlying network
• Why?
• Can improve reliability
• If link from B->G fails, can take few minutes for Internet to recover (meanwhile app can respond in milliseconds to create new path)
• Disseminate data in a scalable way
• Avoid censorship
• Relative delay penalty (aka “Stretch”)
• Ratio of delay in overlay vs. underlying network
APP-LAYER OVERLAY EXAMPLE
• 8/1
6/4/2018
7
OUTLINE
4. Power and energy
Google Facebook
120+ million users
1.15 billion users
• Flexible service management
6/4/2018
9
• Provider licenses applications to users as a service
• e.g., customer relationship management, email, …
• Avoid costs of installation, maintenance, patches, …
• Platform as a Service (Paas)
• Provider offers software platform for building applications
• e.g., Google’s App-Engine
• Infrastructure as a Service (Iaas)
• Provider offers raw computing, storage, and network
• e.g., Amazon’s Elastic Computing Cloud (EC2)
• Avoid buying servers and estimating resource needs
NOT JUST A COLLECTION OF SERVERS
• A data center isn’t just a “small internet”
• Why?
• Trusted administrators
• Except for traffic to/from users
• No need for international standards bodies
• Though why do standards help?
6/4/2018
10
OUTLINE
4. Power and energy
• Data sizes driven by the content that users actually consume
– Growth largely due to higher bitrate content (IP TV/movies, iPhone Facetime)
• Mobile Internet source of new users
• Often constrained by the “last mile”
Wide-area Internet
Data center
• Key differentiator determining success
• Sorting / Searching
• Collaborative Filtering
DATA-INTENSIVE APPLICATION REQUIREMENTS
• Need high “bisection” bandwidth
• Need low variance
• 10/40 Gbps at TOR (and soon endhosts)
• Congestion-free operation/low queuing
FROM NETWORKS TO BACKPLANES
• Massive computing = tightly coupled supercomputer
• Proprietary interconnects
• Data-intensive, web-scale
6/4/2018
14
OUTLINE
4. Power and energy
Network switch
6/4/2018
16
• VM can migrate from one computer to another
VMM VIRTUAL SWITCHES
HARDWARE COMPARISONS
Barroso and Holzle, The Data Center as a Computer: An introduction to the Design of Warehouse-Scale Machines
6/4/2018
21
BIG COMPUTER VS. LOTS-OF-SMALL-COMPUTERS
Barroso and Holzle, The Data Center as a Computer: An introduction to the Design of Warehouse-Scale Machines
BIG COMPUTER VS. LOTS-OF-SMALL-COMPUTERS
Barroso and Holzle, The Data Center as a Computer: An introduction to the Design of Warehouse-Scale Machines
6/4/2018
22
OUTLINE
4. Power and energy
Power provisioning for a
floor tiles
Liquid supply
QUANTIFYING ENERGY-EFFICIENCY: PUE
• E.g., 1 watt of computing meant 1 Watt of cooling!
PUE = (Facility Power) / (Computing Equipment power)
6/4/2018
24
LBNL PUE SURVEY (2007)
Barroso and Holzle, The Data Center as a Computer: An introduction to the Design of Warehouse-Scale Machines
LBNL PUE Survey (2013)
BREAKDOWN OF DATA CENTER OVERHEADS
Barroso and Holzle, The Data Center as a Computer: An introduction to the Design of Warehouse-Scale Machines
NUMBERS FROM JAMES HAMILTON (MSFT, AMAZON)
6/4/2018
26
• Google’s “Chiller- less” data center in Belgium
• Most of the year it is cool enough to not need cooling
• What about on hot days?
• Shed load to other data centers!
OUTLINE
4. Power and energy
POWER-PROPORTIONAL HUMANS
Barroso and Holzle, The Data Center as a Computer: An introduction to the Design of Warehouse-Scale Machines
WEB-SERVICE LOAD FLUCTUATIONS
Barroso and Holzle, The Data Center as a Computer: An introduction to the Design of Warehouse-Scale Machines
6/4/2018
28
DO DIFFERENT COMPONENTS SCALE SIMILARLY?
Barroso and Holzle, The Data Center as a Computer: An introduction to the Design of Warehouse-Scale Machines
CPU UTILIZATION
Barroso and Holzle, The Data Center as a Computer: An introduction to the Design of Warehouse-Scale Machines
6/4/2018
29
POWER USAGE VS. UTILIZATION
OUTLINE
4. Power and energy
CR CR
~ 1,000 servers/pod == IP
CR CR
ATTRIBUTION • These slides are released under an Attribution-NonCommercial-ShareAlike 3.0
Unported (CC BY-NC-SA 3.0) Creative Commons license
• These slides incorporate material from:
• The Datacenter as a Computer: An Introduction to the Design of Warehouse-Scale Machines, 2nd ed., by Barroso, Clidaras, and Hölzle.
6/4/2018
2
ANNOUNCEMENTS (1/2) Please read The Datacenter as a Computer: An Introduction to the Design of Warehouse-Scale Machines, 2nd ed., by Barroso, Clidaras, and Hölzle.
Ideally all of it, but at least Chapters 1-5
Sample finals posted to Piazza’s “resources” page
“Minimal” Homework 5 posted
PROJECT 2 ANNOUNCEMENTS (2/2)
Hash function: use provided function (pass bytes into it instead of string)
Distributed getversion test removed
Download text file, open with text editor to make sure it is ok
Client to/from MDS: we will use our own client (make sure to follow RPC given, don’t create your own)
Will manually look for 2PC implementation (60% not implementing 2PC…)
6/4/2018
3
• Pages 221-224
• Data flows along tree, not underlying network
• Why?
• Can improve reliability
• If link from B->G fails, can take few minutes for Internet to recover (meanwhile app can respond in milliseconds to create new path)
• Disseminate data in a scalable way
• Avoid censorship
• Relative delay penalty (aka “Stretch”)
• Ratio of delay in overlay vs. underlying network
APP-LAYER OVERLAY EXAMPLE
• 8/1
6/4/2018
7
OUTLINE
4. Power and energy
Google Facebook
120+ million users
1.15 billion users
• Flexible service management
6/4/2018
9
• Provider licenses applications to users as a service
• e.g., customer relationship management, email, …
• Avoid costs of installation, maintenance, patches, …
• Platform as a Service (Paas)
• Provider offers software platform for building applications
• e.g., Google’s App-Engine
• Infrastructure as a Service (Iaas)
• Provider offers raw computing, storage, and network
• e.g., Amazon’s Elastic Computing Cloud (EC2)
• Avoid buying servers and estimating resource needs
NOT JUST A COLLECTION OF SERVERS
• A data center isn’t just a “small internet”
• Why?
• Trusted administrators
• Except for traffic to/from users
• No need for international standards bodies
• Though why do standards help?
6/4/2018
10
OUTLINE
4. Power and energy
• Data sizes driven by the content that users actually consume
– Growth largely due to higher bitrate content (IP TV/movies, iPhone Facetime)
• Mobile Internet source of new users
• Often constrained by the “last mile”
Wide-area Internet
Data center
• Key differentiator determining success
• Sorting / Searching
• Collaborative Filtering
DATA-INTENSIVE APPLICATION REQUIREMENTS
• Need high “bisection” bandwidth
• Need low variance
• 10/40 Gbps at TOR (and soon endhosts)
• Congestion-free operation/low queuing
FROM NETWORKS TO BACKPLANES
• Massive computing = tightly coupled supercomputer
• Proprietary interconnects
• Data-intensive, web-scale
6/4/2018
14
OUTLINE
4. Power and energy
Network switch
6/4/2018
16
• VM can migrate from one computer to another
VMM VIRTUAL SWITCHES
HARDWARE COMPARISONS
Barroso and Holzle, The Data Center as a Computer: An introduction to the Design of Warehouse-Scale Machines
6/4/2018
21
BIG COMPUTER VS. LOTS-OF-SMALL-COMPUTERS
Barroso and Holzle, The Data Center as a Computer: An introduction to the Design of Warehouse-Scale Machines
BIG COMPUTER VS. LOTS-OF-SMALL-COMPUTERS
Barroso and Holzle, The Data Center as a Computer: An introduction to the Design of Warehouse-Scale Machines
6/4/2018
22
OUTLINE
4. Power and energy
Power provisioning for a
floor tiles
Liquid supply
QUANTIFYING ENERGY-EFFICIENCY: PUE
• E.g., 1 watt of computing meant 1 Watt of cooling!
PUE = (Facility Power) / (Computing Equipment power)
6/4/2018
24
LBNL PUE SURVEY (2007)
Barroso and Holzle, The Data Center as a Computer: An introduction to the Design of Warehouse-Scale Machines
LBNL PUE Survey (2013)
BREAKDOWN OF DATA CENTER OVERHEADS
Barroso and Holzle, The Data Center as a Computer: An introduction to the Design of Warehouse-Scale Machines
NUMBERS FROM JAMES HAMILTON (MSFT, AMAZON)
6/4/2018
26
• Google’s “Chiller- less” data center in Belgium
• Most of the year it is cool enough to not need cooling
• What about on hot days?
• Shed load to other data centers!
OUTLINE
4. Power and energy
POWER-PROPORTIONAL HUMANS
Barroso and Holzle, The Data Center as a Computer: An introduction to the Design of Warehouse-Scale Machines
WEB-SERVICE LOAD FLUCTUATIONS
Barroso and Holzle, The Data Center as a Computer: An introduction to the Design of Warehouse-Scale Machines
6/4/2018
28
DO DIFFERENT COMPONENTS SCALE SIMILARLY?
Barroso and Holzle, The Data Center as a Computer: An introduction to the Design of Warehouse-Scale Machines
CPU UTILIZATION
Barroso and Holzle, The Data Center as a Computer: An introduction to the Design of Warehouse-Scale Machines
6/4/2018
29
POWER USAGE VS. UTILIZATION
OUTLINE
4. Power and energy
CR CR
~ 1,000 servers/pod == IP
CR CR
Related Documents
