Cloud computing
•Transferir como PPTX, PDF•
0 gostou•482 visualizações
Presented Qatar Foundation Annual Research Forum, Nov. 2011
Recomendados
Recomendados
Mais conteúdo relacionado
Mais procurados
Mais procurados (19)
Destaque
Destaque (7)
Semelhante a Cloud computing
Semelhante a Cloud computing (20)
Mais de Lew Tucker
Mais de Lew Tucker (12)
Último
Último (20)
Cloud computing
- 1. Cloud Computing Lew Tucker, Ph.D. VP/CTO Cloud Computing Cisco Systems, Inc QCR Research Challenges – Nov 2011 - Qatar
- 2. © 2010 Cisco and/or its affiliates. All rights reserved. 2
- 3. Adoption Curve Traditional Cloud Computing Data Centers Public or Private 2000 2005 2010 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . © 2010 Cisco and/or its affiliates. All rights reserved. 3
- 4. © 2010 Cisco and/or its affiliates. All rights reserved. 4
- 5. “New model that provides on- demand access to applications and computing resources over the internet” - e.g. in the cloud © 2010 Cisco and/or its affiliates. All rights reserved. 5
- 6. per Virtual Machine 10 servers for 4 days == 1000 servers for an hour © 2010 Cisco and/or its affiliates. All rights reserved. 6
- 7. Gordon Moore Adam Smith © 2010 Cisco and/or its affiliates. All rights reserved. 7
- 8. “I really don’t want to operate data centers anymore… We’d rather spend our time giving our customers great service and writing great software” – Don MacAskill, CEO, Smugmug © 2010 Cisco and/or its affiliates. All rights reserved. 8
- 9. Complexity Grows With Number of Apps Corp Mktg Finance Engineering HR App App App App App App App OS OS OS OS OS OS OS Physical Physical Physical Physical Physical Physical Physical Server Server Server Server Server Server Server DB DB DB Storage DB DB Poor Utilization Inflexible Infrastructure © 2010 Cisco and/or its affiliates. All rights reserved. 9
- 10. Separate Infrastructure from the Applications Corp Mktg Finance Engineering HR App App App App App App App OS OS OS OS OS OS OS Virtual Virtual Virtual Virtual Virtual Virtual Virtual Machine Machine Machine Machine Machine Machine Machine Cloud Infrastructure Service Cloud Infrastructure Service Physical Server Physical DB Service Queue Server Storage Storage Physical Server Physical Server © 2010 Cisco and/or its affiliates. All rights reserved. 10
- 11. Corp Mktg Finance Engineering HR App App App App App App App OS OS OS OS OS OS OS Virtual Virtual Virtual Virtual Virtual Virtual Virtual Machine Machine Machine Machine Machine Machine Machine API-driven services Self-service portal Selective application mgmt Cloud Infrastructure Service Pool of shared resources Physical Server Physical DB Service Queue Server Storage Storage Physical Server Physical Server © 2010 Cisco and/or its affiliates. All rights reserved. 11
- 12. Corp Mktg Finance Engineering HR App App App App App App App OS OS OS OS OS OS OS Virtual Virtual Virtual Virtual Virtual Virtual Virtual Machine Machine Machine Machine Machine Machine Machine © 2010 Cisco and/or its affiliates. All rights reserved. 12
- 13. App App App OS OS App App App OS OS OS Virtual Virtual Virtual OS Machine Machine Machin e Virtual Virtual Virtual Machine Machine Machine App App App App App App App OS OS OS OS OS OS OS App Virtual Virtual Virtual Virtual Virtual Virtual Virtual Machine Machine App Machine Machine Machine Machine Machine OS App App OS OS OS Virtual Machine Virtual Virtual Virtual Machine Machine Machine © 2010 Cisco and/or its affiliates. All rights reserved. 13
- 14. Connection Machine (65,000 processors) Cray XMP © 2010 Cisco and/or its affiliates. All rights reserved. 14
- 15. © 2010 Cisco and/or its affiliates. All rights reserved. 15
- 16. © 2010 Cisco and/or its affiliates. All rights reserved. 16
- 17. © 2010 Cisco and/or its affiliates. All rights reserved. 17
- 18. 2,000 servers, 240 teraFLOPS Rent your own Top500 supercomputer for $1000 per hour 290 Instances (8 cores, 60 GB RAM, 3.4 TB local storage) 63.7 teraFlops – Petabyte storage Even less on Spot market © 2010 Cisco and/or its affiliates. All rights reserved. 18
- 19. What we can learn from Internet Companies Web Enterprise Approach Approach Scale-out Vertical scaling Architecture HA failover model Design for Failure Transactional Information- centric Application specific Commodity Infrastructure systems Commercial Open Source Software Innovation and SCALE © 2011 Cisco and/or its affiliates. All rights reserved. 19
- 20. Google’s data center on the Columbia river, Oregon © 2010 Cisco and/or its affiliates. All rights reserved. 20
- 21. MapReduce/Hadoop Documents Map workers: reads Map Map Map input, invoking Map() function Map which finds words and outputs <word, cnt> <word, cnt> <word, cnt> <word, cnt> records for reduce workers hash(word, mod R) P1 P1 P1 P1 Reduce workers: reads sorted Map output, Reduce Reduce Reduce invoking Reduce() task which <word, cnt> <word, cnt> <word, cnt> sums counts for each word Result Result Result <apple, 2,045,455> <house, 111,341> <cat, 1,033,746> <barn, 254,345> <kitchen, 4,678> <horse, 25,387> © 2010 Cisco and/or its affiliates. All rights reserved. 21
- 22. Virtual Servers 4000 3000 2000 1500 50 to 3500 servers over a 3 day period!! 1000 500 50 Animoto.com case study 4/13 4/13 4/14 4/15 4/15 4/16 4/17 4/17 4/18 © 2010 Cisco and/or its affiliates. All rights reserved. 22
- 23. • OpenStack is open source software that allows any company to build their own public or private cloud. • Turns data center infrastructure into a service • Think of it as: “software that powers Amazon’s EC2+S3”. © 2011 Cisco and/or its affiliates. All rights reserved. 23
- 24. Government Media Financial Services Healthcare Games Pharma Seamlessly Connected Public Private Securely Accessed © 2011 Cisco and/or its affiliates. All rights reserved. 24
- 25. Data Centers Mobile Devices Sensors Data Center Reference: J. Rabaey, “A Brand New Wireless Day,” Keynote Presentation, ASPDAC Jan. 2008 © 2010 Cisco and/or its affiliates. All rights reserved. 25
- 26. Data Here Data Here Data Here Data Here Public Data Sets, Live Streams, Cloud-based Tools, and Services © 2010 Cisco and/or its affiliates. All rights reserved. 26
- 27. In the internet age, whatever platforms have the most users, data, and developers win © 2010 Cisco and/or its affiliates. All rights reserved. 27
Notas do Editor
- Cloud computing isn’t so much about fundamentally new technology, but rather a change in the way that it is delivered. What is interesting is that it is driven not only by the advancement of technology, as characterized by Moore’s law (cost performance doubling every 18 months), but also by economics. The market for consumer electronics has driven down the cost of processors, memory, and disk storage such that computing is becoming relatively cheap.Even more interesting is the fact that services like Amazon’s AWS and others are setting a market price for computing. Today an hour computing costs around 10 cents per hour, and the cost to store a gigabyte for a month is around 15 cents. With the new infrastructure-as-a-service delivery model, one can also pay only for what you use and avoid the traditional capital investment corporations have made in IT.Most importantly, from a historical perspective we have for the first time an easily measured, market-determined cost of computing.
- So what is it about the cloud computing that makes it so much more desirable as a model for how data centers are run? Here we have a simplified model of an enterprise IT organization supporting multiple departments. This same approach works for managed service providers if you imagine the departments are actually different customers or tenants.Watch what happens when the data center grows as more applications are added. As each new application is added, a new system configuration is created consisting of a server, operating system, storage, and the app itself. As more applications are added, the system grows in complexity. There is very little scaling advantage. Because each application setup took getting approval, and months of development, even those apps that are only used occasionally never leave for fear the department wouldn’t ever get the servers back again. This leads to the so-called “server hugging” and the resulting poor utilization and inflexible infrastructure.(While we are showing this for an enterprise, the same holds true for a managed services provider by viewing each application as a different tenant.)
- In the cloud computing model things are different. In this case, the services organization provides a cloud infrastructure service, on top of which applications are deployed fully configured with their own operating system and configuration. We have de-coupled the complexity of applications from the underlying infrastructure. The application layer and the infrastructure layer are managed independently.