STANFORD UNIVERSITY • INFORMATION TECHNOLOGY SERVICES
Clearing Away the Clouds: Cloud Computing at Stanford
Tech Briefing Turing Auditorium, 1/29/2010
Phil Reese, IT Services
STANFORD UNIVERSITY • INFORMATION TECHNOLOGY SERVICES
1/29/10 Clearing Away the Clouds: Cloud Computing at Stanford page 1
Topics:
A walk down memory lane, which leads to the clouds.
A chair with four legs.
Several cloudy examples.
Stanford’s forecast, sunny with a likelihood of clouds.
STANFORD UNIVERSITY • INFORMATION TECHNOLOGY SERVICES
1/29/10 Clearing Away the Clouds: Cloud Computing at Stanford page 2
Several Scenarios or Use Cases:
1- Startup looking to put up an ecommerce web site, large data base application.
2- Development group wanting to test next version of code before putting it into production.
3- Researchers having to do large computations after a physics experiment.
4- Multi university researchers needing to do very large computations.
5- Many different researchers with lots of different computations trying to leverage the same computing resources.
STANFORD UNIVERSITY • INFORMATION TECHNOLOGY SERVICES
1/29/10 Clearing Away the Clouds: Cloud Computing at Stanford page 3
Scenario #1:
Startup looking to put up an ecommerce web site, large data base application. • Forest Gump Chocolate ecommerce site (circa 1996)
• Started with a single large x86 server hosted in the startups’ office.
STANFORD UNIVERSITY • INFORMATION TECHNOLOGY SERVICES
1/29/10 Clearing Away the Clouds: Cloud Computing at Stanford page 4
Scenario #1:
Startup looking to put up an ecommerce web site, large data base application. • Forest Gump Chocolate ecommerce site (circa 1996)
• Started with a single large x86 server hosted in the startups’ office.
• Moved to a collocation facility to improve bandwidth availability.
STANFORD UNIVERSITY • INFORMATION TECHNOLOGY SERVICES
1/29/10 Clearing Away the Clouds: Cloud Computing at Stanford page 5
Scenario #1:
Startup looking to put up an ecommerce web site, large data base application. • Forest Gump Chocolate ecommerce site (circa 1996)
• Started with a single large x86 server hosted in the startups’ office.
• Moved to a collocation facility to improve bandwidth availability.
• Interface slowness during peak times.
STANFORD UNIVERSITY • INFORMATION TECHNOLOGY SERVICES
1/29/10 Clearing Away the Clouds: Cloud Computing at Stanford page 6
Scenario #1:
Startup looking to put up an ecommerce web site, large data base application. • Forest Gump Chocolate ecommerce site (circa 1996)
• Started with a single large x86 server hosted in the startups’ office.
• Moved to a collocation facility to improve bandwidth availability.
• Interface slowness during peak times. • Processing slow downs under load.
STANFORD UNIVERSITY • INFORMATION TECHNOLOGY SERVICES
1/29/10 Clearing Away the Clouds: Cloud Computing at Stanford page 7
Scenario #1:
Startup looking to put up an ecommerce web site, large data base application. • Forest Gump Chocolate ecommerce site (circa 1996)
• Started with a single large x86 server hosted in the startups’ office.
• Moved to a collocation facility to improve bandwidth availability.
• Interface slowness during peak times. • Processing slow downs under load. • Final system setup before being acquired by See’s Chocolate.
STANFORD UNIVERSITY • INFORMATION TECHNOLOGY SERVICES
1/29/10 Clearing Away the Clouds: Cloud Computing at Stanford page 8
Scenario #1:
Startup looking to put up an ecommerce web site, large data base application. • See’s Chocolate ecommerce site (circa 2001)
• Several web servers in front of an application and data base server.
STANFORD UNIVERSITY • INFORMATION TECHNOLOGY SERVICES
1/29/10 Clearing Away the Clouds: Cloud Computing at Stanford page 9
Scenario #1:
Startup looking to put up an ecommerce web site, large data base application. • See’s Chocolate ecommerce site (circa 2001)
• Several web servers in front of an application and data base server.
• Running too slow, so the application and data base portions were separated to dedicated servers.
STANFORD UNIVERSITY • INFORMATION TECHNOLOGY SERVICES
1/29/10 Clearing Away the Clouds: Cloud Computing at Stanford page 10
Scenario #1:
Startup looking to put up an ecommerce web site, large data base application. • See’s Chocolate ecommerce site (circa 2001)
• Several web servers in front of an application and data base server.
• Running too slow, so the application and data base portions were separated to dedicated servers.
• Still running too slow, so multiple servers at every level.
STANFORD UNIVERSITY • INFORMATION TECHNOLOGY SERVICES
1/29/10 Clearing Away the Clouds: Cloud Computing at Stanford page 11
Scenario #1:
Startup looking to put up an ecommerce web site, large data base application. • See’s Chocolate ecommerce site (circa 2001)
• Several web servers in front of an application and data base server.
• Running too slow, so the application and data base portions were separated to dedicated servers.
• Still running too slow, so multiple servers at every level. • Works but required complex networking setup.
STANFORD UNIVERSITY • INFORMATION TECHNOLOGY SERVICES
1/29/10 Clearing Away the Clouds: Cloud Computing at Stanford page 12
Scenario #1:
Startup looking to put up an ecommerce web site, large data base application. • Wonka’s Chocolate ecommerce site (circa 2009)
• In addition to dedicated hardware, the site now uses cloud based resources to meet peak demands at all three layers of the services stack.
STANFORD UNIVERSITY • INFORMATION TECHNOLOGY SERVICES
1/29/10 Clearing Away the Clouds: Cloud Computing at Stanford page 13
Scenario #1:
Startup looking to put up an ecommerce web site, large data base application. • Wonka’s Chocolate ecommerce site (circa 2009)
• In addition to dedicated hardware, the site now uses cloud based resources to meet peak demands at all three layers of the services stack.
• Key feature of this successful site is CLUSTERING and Cloud.
STANFORD UNIVERSITY • INFORMATION TECHNOLOGY SERVICES
1/29/10 Clearing Away the Clouds: Cloud Computing at Stanford page 14
Scenario #3:
Researchers having to do large computations after a physics experiment. (circa 1997) • Since desktop computers were pretty slow and underpowered at
that time, the researcher would look to central computing resources.
STANFORD UNIVERSITY • INFORMATION TECHNOLOGY SERVICES
1/29/10 Clearing Away the Clouds: Cloud Computing at Stanford page 15
Scenario #3:
Researchers having to do large computations after a physics experiment. (circa 1997) • Since desktop computers were pretty slow and underpowered at
that time, the researcher would look to central computing resources.
• These would typically be very expensive and high end special purpose computers. Names like Cray, Convex, Masspar, IBM, and HP were typical.
STANFORD UNIVERSITY • INFORMATION TECHNOLOGY SERVICES
1/29/10 Clearing Away the Clouds: Cloud Computing at Stanford page 16
Scenario #3:
Researchers having to do large computations after a physics experiment. (circa 1997) • Since desktop computers were pretty slow and underpowered at
that time, the researcher would look to central computing resources.
• These would typically be very expensive and high end special purpose computers. Names like Convex, Masspar, IBM, and HP were typical.
• If your university didn’t have these resources, you might be able to get some time on similar machines run at a few central places, called SuperComputing Centers, across the US.
STANFORD UNIVERSITY • INFORMATION TECHNOLOGY SERVICES
1/29/10 Clearing Away the Clouds: Cloud Computing at Stanford page 17
Scenario #3:
Researchers having to do large computations after a physics experiment. (circa 1997) • Since desktop computers were pretty slow and underpowered at
that time, the researcher would look to central computing resources.
• These would typically be very expensive and high end special purpose computers. Names like Convex, Masspar, IBM, and HP were typical.
• If your university didn’t have these resources, you might be able to get some time on similar machines run at a few central places, called SuperComputing Centers, across the US.
• As these high priced systems became out of reach and Linux and better x86 systems started to be available, clusters of x86 machines began replacing the special purpose supercomputers.
STANFORD UNIVERSITY • INFORMATION TECHNOLOGY SERVICES
1/29/10 Clearing Away the Clouds: Cloud Computing at Stanford page 18
Scenario #3:
Researchers having to do large computations after a physics experiment. (circa 1997) • Since desktop computers were pretty slow and underpowered at
that time, the researcher would look to central computing resources.
• These would typically be very expensive and high end special purpose computers. Names like Convex, Masspar, IBM, and HP were typical.
• If your university didn’t have these resources, you might be able to get some time on similar machines run at a few central places, called SuperComputing Centers, across the US.
• As these high priced systems became out of reach and Linux and better x86 systems started to be available, clusters of x86 machines began replacing the special purpose supercomputers.
• The beginning of the GRID computing concept.
STANFORD UNIVERSITY • INFORMATION TECHNOLOGY SERVICES
1/29/10 Clearing Away the Clouds: Cloud Computing at Stanford page 19
Scenario #4:
Multi university researchers needing to do very large computations. (circa 2003) • The grid helped individual researchers get the computing resources
needed for their work. Multiple researchers needing to work on the same data found the Grid difficult to use but was the only option.
STANFORD UNIVERSITY • INFORMATION TECHNOLOGY SERVICES
1/29/10 Clearing Away the Clouds: Cloud Computing at Stanford page 20
Scenario #4:
Multi university researchers needing to do very large computations. (circa 2003) • The grid helped individual researchers get the computing resources
needed for their work. Multiple researchers needing to work on the same data found the Grid difficult to use but was the only option.
• Lots of effort and energy put forth for national computing centers to address this issue, since it was happening all over the country.
STANFORD UNIVERSITY • INFORMATION TECHNOLOGY SERVICES
1/29/10 Clearing Away the Clouds: Cloud Computing at Stanford page 21
Scenario #4:
Multi university researchers needing to do very large computations. (circa 2003) • The grid helped individual researchers get the computing resources
needed for their work. Multiple researchers needing to work on the same data found the Grid difficult to use but was the only option.
• Lots of effort and energy put forth for national computing centers to address this issue, since it was happening all over the country.
• There is always reluctance to centralizing resources when it assumes that limitations will be put on the use of those resources. • Version of Linux • Library availability • Different types of science needed different applications
STANFORD UNIVERSITY • INFORMATION TECHNOLOGY SERVICES
1/29/10 Clearing Away the Clouds: Cloud Computing at Stanford page 22
Two over the four legs:
Clusters of computers at Universities Grids were formed of the clusters so that ever more
computing power could be grouped together. • Sounds good! What were the issues?
STANFORD UNIVERSITY • INFORMATION TECHNOLOGY SERVICES
1/29/10 Clearing Away the Clouds: Cloud Computing at Stanford page 23
Two over the four legs:
Clusters of computers at Universities Grids were formed of the clusters so that ever more
computing power could be grouped together. • Sounds good! What were the issues?
• Staying in sync with your own cluster and the corollary, staying in sync with the grid you belonged to.
STANFORD UNIVERSITY • INFORMATION TECHNOLOGY SERVICES
1/29/10 Clearing Away the Clouds: Cloud Computing at Stanford page 24
Two over the four legs:
Clusters of computers at Universities Grids were formed of the clusters so that ever more
computing power could be grouped together. • Sounds good! What were the issues?
• Staying in sync with your own cluster and the corollary, staying in sync with the grid you belonged to.
• What version of Linux was in most favor at the moment?
STANFORD UNIVERSITY • INFORMATION TECHNOLOGY SERVICES
1/29/10 Clearing Away the Clouds: Cloud Computing at Stanford page 25
Two of the four legs:
Clusters of computers at Universities Grids were formed of the clusters so that ever more
computing power could be grouped together. • Sounds good! What were the issues?
• Staying in sync with your own cluster and the corollary, staying in sync with the grid you belonged to.
• What version of Linux was in most favor at the moment? • How to coordinate the best use of the grid? The development of
the GLOBUS Tool Kit.
STANFORD UNIVERSITY • INFORMATION TECHNOLOGY SERVICES
1/29/10 Clearing Away the Clouds: Cloud Computing at Stanford page 26
Scenario #2:
Development group wanting to test next version of code before putting it into production. (circa 2007) • At last a simple answer, just get a matching server to the production
box and test away!
STANFORD UNIVERSITY • INFORMATION TECHNOLOGY SERVICES
1/29/10 Clearing Away the Clouds: Cloud Computing at Stanford page 27
Scenario #2:
Development group wanting to test next version of code before putting it into production. (circa 2007) • At last a simple answer, just get a matching server to the production
box and test away! • Oh, the boss says that will cost too much and not be used
enough to justify the cost.
STANFORD UNIVERSITY • INFORMATION TECHNOLOGY SERVICES
1/29/10 Clearing Away the Clouds: Cloud Computing at Stanford page 28
Scenario #2:
Development group wanting to test next version of code before putting it into production. (circa 2007) • At last a simple answer, just get a matching server to the production
box and test away! • Oh, the boss says that will cost too much and not be used
enough to justify the cost. • Borrow your neighbors system to run the tests. With a collection of
three or groups, the purchase of a spare test system makes more sense.
STANFORD UNIVERSITY • INFORMATION TECHNOLOGY SERVICES
1/29/10 Clearing Away the Clouds: Cloud Computing at Stanford page 29
Scenario #2:
Development group wanting to test next version of code before putting it into production. (circa 2007) • At last a simple answer, just get a matching server to the production
box and test away! • Oh, the boss says that will cost too much and not be used
enough to justify the cost. • Borrow your neighbors system to run the tests. With a collection of
three or groups, the purchase of a space test system makes more sense. • Darn, two of the four groups are on the same production refresh
cycle and need the box at the same time!
STANFORD UNIVERSITY • INFORMATION TECHNOLOGY SERVICES
1/29/10 Clearing Away the Clouds: Cloud Computing at Stanford page 30
Scenario #2:
Development group wanting to test next version of code before putting it into production. (circa 2007) • At last a simple answer, just get a matching server to the production
box and test away! • Oh, the boss says that will cost too much and not be used
enough to justify the cost. • Borrow your neighbors system to run the tests. With a collection of
three or groups, the purchase of a space test system makes more sense. • Darn, two of the four groups are on the same production refresh
cycle and need the box at the same time! • Use a VIRTUAL machine on that box so each group can have their
own configuration and not interfere with the other groups.
STANFORD UNIVERSITY • INFORMATION TECHNOLOGY SERVICES
1/29/10 Clearing Away the Clouds: Cloud Computing at Stanford page 31
Three of the four legs:
Clusters of computers at Universities Grids were formed of the clusters so that ever more
computing power could be grouped together. Virtual machines allow a single machine to be used
for multiple functions. • Better use of idle resources • Begins to shape a framework for standardizing system OS’s and
machine configurations. • Virtualizes the machine so that it can easily move to different
hardware architectures.
STANFORD UNIVERSITY • INFORMATION TECHNOLOGY SERVICES
1/29/10 Clearing Away the Clouds: Cloud Computing at Stanford page 32
Scenario #5:
Many different researchers with lots of different computations trying to leverage the same computing resources. • Clusters work great for individual researchers at individual
Universities. • Deeper look at how clusters actually operate.
STANFORD UNIVERSITY • INFORMATION TECHNOLOGY SERVICES
1/29/10 Clearing Away the Clouds: Cloud Computing at Stanford page 33
Scenario #5:
Many different researchers with lots of different computations trying to leverage the same computing resources. • Clusters work great for individual researchers at individual
Universities. • Deeper look at how clusters actually operate.
• Grids have a pretty long history but never really made the grade since they required specific versions of code and even hardware to some extent, though they got the world thinking about sharing facilities that might have some idle time.
STANFORD UNIVERSITY • INFORMATION TECHNOLOGY SERVICES
1/29/10 Clearing Away the Clouds: Cloud Computing at Stanford page 34
Scenario #5:
Many different researchers with lots of different computations trying to leverage the same computing resources. • Clusters work great for individual researchers at individual
Universities. • Deeper look at how clusters actually operate.
• Grids have a pretty long history but never really made the grade since they required specific versions of code and even hardware to some extent, though they got the world thinking about sharing facilities that might have some idle time.
• The answer to this turns out the be the CLOUD.
STANFORD UNIVERSITY • INFORMATION TECHNOLOGY SERVICES
1/29/10 Clearing Away the Clouds: Cloud Computing at Stanford page 35
Four of the four legs:
Clusters of computers at Universities Grids were formed of the clusters so that ever more
computing power could be grouped together. Virtual machines allow a single machine to be used
for multiple functions. Cloud computing
• Arguably, Cloud computing is a combination of virtual machines, clusters of computers, grids of clusters and several additional features.
STANFORD UNIVERSITY • INFORMATION TECHNOLOGY SERVICES
1/29/10 Clearing Away the Clouds: Cloud Computing at Stanford page 36
Four of the four legs:
Clusters of computers at Universities Grids were formed of the clusters so that ever more
computing power could be grouped together. Virtual machines allow a single machine to be used
for multiple functions. Cloud computing
• Arguably, Cloud computing is a combination of virtual machines, clusters of computers, grids of clusters and several additional features. • Types of clouds:
• Private and public
STANFORD UNIVERSITY • INFORMATION TECHNOLOGY SERVICES
1/29/10 Clearing Away the Clouds: Cloud Computing at Stanford page 37
Four of the four legs:
Clusters of computers at Universities Grids were formed of the clusters so that ever more
computing power could be grouped together. Virtual machines allow a single machine to be used
for multiple functions. Cloud computing
• Arguably, Cloud computing is a combination of virtual machines, clusters of computers, grids of clusters and several additional features. • Types of clouds:
• Private and public • Stats and growth
STANFORD UNIVERSITY • INFORMATION TECHNOLOGY SERVICES
1/29/10 Clearing Away the Clouds: Cloud Computing at Stanford page 38
Several cloudy examples.
Fun in the clouds! • Types of applications that are ripe for cloud use.
• Peak loads • Testing/prototyping • Parallel code
STANFORD UNIVERSITY • INFORMATION TECHNOLOGY SERVICES
1/29/10 Clearing Away the Clouds: Cloud Computing at Stanford page 39
Several cloudy examples.
Fun in the clouds! • Types of applications that are ripe for cloud use.
• Peak loads • Testing/prototyping • Parallel code
• Clouds just aren’t for computing. • Turns out most applications require compute cycles AND
data, typically from a storage system.
STANFORD UNIVERSITY • INFORMATION TECHNOLOGY SERVICES
1/29/10 Clearing Away the Clouds: Cloud Computing at Stanford page 40
Several cloudy examples.
Fun in the clouds! • Types of applications that are ripe for cloud use.
• Peak loads • Testing/prototyping • Parallel code
• Clouds just aren’t for computing. • Turns out most applications require compute cycles AND
data, typically from a storage system. • Many public and private clouds include CPU and
STORAGE as the full offering.
STANFORD UNIVERSITY • INFORMATION TECHNOLOGY SERVICES
1/29/10 Clearing Away the Clouds: Cloud Computing at Stanford page 41
Several cloudy examples.
Fun in the clouds! • Types of applications that are ripe for cloud use.
• Peak loads • Testing/prototyping • Parallel code
• Clouds just aren’t for computing. • Turns out most applications require compute cycles AND
data, typically from a storage system. • Many public and private clouds include CPU and
STORAGE as the full offering. • Amazon is the poster child!
• http://aws.amazon.com/
STANFORD UNIVERSITY • INFORMATION TECHNOLOGY SERVICES
1/29/10 Clearing Away the Clouds: Cloud Computing at Stanford page 42
Several cloudy examples.
Creating accounts for both EC2 and S3 • Follow the instructions… • Open an instance and web page • Open Cyberduck
STANFORD UNIVERSITY • INFORMATION TECHNOLOGY SERVICES
1/29/10 Clearing Away the Clouds: Cloud Computing at Stanford page 43
Several cloudy examples.
Creating accounts for both EC2 and S3 • Follow the instructions… • Open an instance and web page • Open Cyberduck • Discussion about instances and coordinating instances and
scheduling.
STANFORD UNIVERSITY • INFORMATION TECHNOLOGY SERVICES
1/29/10 Clearing Away the Clouds: Cloud Computing at Stanford page 44
Several cloudy examples.
Creating accounts for both EC2 and S3 • Follow the instructions… • Open an instance and web page • Open Cyberduck • Discussion about instances and coordinating instances and
scheduling.
There must be more interesting apps out there • https://rightsignature.com/ • http://www.dropbox.com/
• Use this to sign up and get an extra 250mb • https://www.dropbox.com/referrals/NTg0MDMyOQ
• http://ylastic.com/
STANFORD UNIVERSITY • INFORMATION TECHNOLOGY SERVICES
1/29/10 Clearing Away the Clouds: Cloud Computing at Stanford page 45
Several cloudy examples.
Creating accounts for both EC2 and S3 • Follow the instructions… • Open an instance and web page • Open Cyberduck • Discussion about instances and coordinating instances and
scheduling.
There must be more interesting apps out there • https://rightsignature.com/ • http://www.dropbox.com/
• Use this to sign up and get an extra 250mb • https://www.dropbox.com/referrals/NTg0MDMyOQ
• http://ylastic.com/
Don’t forget Google, Hadoop, and MapReduce
STANFORD UNIVERSITY • INFORMATION TECHNOLOGY SERVICES
1/29/10 Clearing Away the Clouds: Cloud Computing at Stanford page 46
Stanford’s forecast, sunny with a likelihood of clouds.
Using AWS over a 10gig link. Options for educational/academic use Private versus public
STANFORD UNIVERSITY • INFORMATION TECHNOLOGY SERVICES
1/29/10 Clearing Away the Clouds: Cloud Computing at Stanford page 47
Stanford’s forecast, sunny with a likelihood of clouds.
Using AWS over a 10gig link. Options for educational/academic use Private versus public
A chair with four five legs. • Distributed computing
• Folding at home • Storage at home