Application Architecture For The Cloud
Download as PPT, PDF14 likes3,598 views
The document discusses considerations for building application architectures for cloud computing. It outlines some benefits and drawbacks of moving applications to the cloud. It also discusses components needed for cloud applications, such as web UIs, agile scaling, live upgrades, and using services like Amazon S3, SimpleDB, and EC2. It proposes an "Apache Cloud Computing Edition" with technologies like Hadoop and MapReduce and addresses needs like configuration, resource management, and testing in the cloud.
Application Architecture For The Cloud
- 1. Application Architecture for the Cloud Steve Loughran Julio Guijarro
- 2. 8 June 2009
- 3. Why move to the cloud? Good Cost Outsourcing hardware problems Move from capital to pay-as-you-go To handle Petabytes of data For a business plan that might work Bad: to avoid the operations team 8 June 2009
- 4. What is a cloud application? The program that is run The code/data needed to run it in a datacentre Anything needed to configure, monitor and manage the system 8 June 2009
- 5. This is not a cloud application 8 June 2009 Enterprise Java on Highly Available servers
- 6. This is not a cloud application: Java EE 8 June 2009
- 7. Things must change! Web UI for users, affiliates, marketing, operations Agile machine management is part of the API Scale up -and down Live upgrade of running system Persistence with key-value stores A Petabyte filesystem is part of the application MapReduce jobs close the loop Developers deploy to the cloud to test 8 June 2009
- 8. 8 June 2009
- 9. EC2 as the cloud provider S3 for persistence, public downloads SimpleDB provides key-value storage, but query costs unpredictable. Typica for EC2 services API http://code.google.com/p/typica/ AWS IP rental or Dyndns for hostnames No image management services 8 June 2009 No standard Apache Stack
- 10. Sun, IBM, HP as the cloud providers S3 -like filestore EC2 and Sun RESTy APIs Unknown queue and keystore services More secure networking? Image management services? 8 June 2009 No standard Apache Stack
- 11. Private cloud Eucalyptus for deploying Xen images Various persistence options Private filestore: HDFS, kfs, Lustre Kickstart for image management? 8 June 2009 No standard Apache Stack
- 12. Whoever owns the API owns the application for its life Whoever owns the data owns you 8 June 2009
- 13. Apache Cloud Computing Edition Diverse mix of high-level technologies Very large filestore at the bottom : Hadoop APIs, Java 7 NIO, Fuse, WebDAV MapReduce phase for post-processing We need stories for : persistence, configuration, resource management 8 June 2009 A standard Apache Stack!
- 14. Front End The existing Web Front ends should work: Servlets, JSP, wicket, PSP, grails (maybe with memcached) Glue: queues, scatter-gather, tuple-space, events Everything needs to handle an agile world Everything needs instrumenting for management 8 June 2009
- 15. Persistence ? 8 June 2009 Cassandra SimpleJPA PNUTS/Sherpa
- 16. Everything needs a REST API REST is the long-haul API -why have a separate internal one? Jersey JAX-RPC is very nice Client API evolving Http Components/HttpClient can be the foundation for the Apache client; needs AWS support. Also: 8 June 2009
- 17. Events and messages "disk 3436 is failing" Bluetooth phone 04:5a:1f:c2:87:91 entered cell 56 in London NW2 Queued purchases with card numbers 8 June 2009 internal and external events: reliability, scalability, triggered actions
- 18. Resource Management? HA resource manager to monitor front end/back end load and request/release machines on demand Kill unhealthy nodes (liveness, performance) Programmable policies (money vs. load) Choreograph live upgrade/migration Resource Manager as a service 8 June 2009
- 19. Configuration & Management LDAP (and APIs) key-value stores SmartFrog moving to Apache license What is Spring planning in this area? 8 June 2009
- 20. Development How to build and test in this world? How to step through a program running on a remote datacentre? How to control testing costs? 8 June 2009 Eclipse won the Java IDE battle - It now needs to compete with Visual Studio + Azure
- 21. Testing -your first terabyte of data 8 June 2009 Lots of opportunities here! protected void map(Text key, Text test, Context context) throws IOException, InterruptedException { TestResult result = new TestResult(); Class<?> testClass = loadClass(context, test); Test testSuite = JUnitMRUtils.extractTest(testClass); TestSuiteRun tsr = new TestSuiteRun(); result.addListener(tsr); testSuite.run(result); for (SingleTestRun singleTestRun : tsr.getTests()) { context.write(new Text(singleTestRun.name), singleTestRun); } }
- 22. Testing -the infrastructure can help 8 June 2009 Pseudo-RNG driven cluster configuration
- 23. Cirrus Cloud Testbed? HP, Intel, Yahoo!, universities Heterogeneous, multiple datacentres Offering datacentre time, not specific apps Low-level API for physical machines Cloud-API for virtual machines Paying customers? No, not yet. Open source projects? I hope so 8 June 2009
- 24. What next? Apache has the core of a Cloud Computing stack How do we take this and: integrate the various pieces? extend them where appropriate? provide an alternative to AppEngine and Azureus? 8 June 2009
- 25. Call to action Stop writing EJB apps Start collecting as much data as you can and feeding that MapReduce mining-phase. Design for: distributed not-quite-Posix filesystem, message queues, name-value databases Apache: let's build our own cloud platform 8 June 2009
- 27. VM Image Management AMI Image sprawl: 10%/month Old images are a security risk The whole PXE+Kickstart process is built for physical machines. This is not an Apache problem, but we'll need to work with the OS vendors & others to integrate 8 June 2009
- 28. HDFS improvements Scale, availability, small files: hierarchical namenodes? Could it be a general purpose media store? For web sites? 8 June 2009
- 29. Amazon EC2 8 June 2009 Host S3 Storage AMI (Xen VM) AMI (Xen VM) /mnt Host AMI (Xen VM) AMI (Xen VM) Public Internet /mnt /mnt /mnt Fast (free) network free access; slow initial read time pay per GET; per megabyte $ $ $ $ $
- 30. 8 June 2009
Editor's Notes
- #2: This is a talk on application architecture for the cloud.