Supercharge Your Applications

  • View

  • Download

Embed Size (px)


1. Pop quiz whos this? Ian Fleming creator of James Bond Second question what was James Bonds first car? 2. It was this (not the Aston Martin DB5 from the 1964 movie Goldfinger)in the first Bond novel, 'Casino Royale' published in 1953 Fleming put our James in one of these. This is a 4.5 litre supercharged Bentley from the 1920s. W.O. Bentley famously said when talking about engine performance - "there's no replacement for displacement". One of his engineers said ah ha, I'll show you, and bolted a supercharger on to the engine of the regular 4.5 litre Bentley - the supercharger is the big silver box on the front of this car - forcing more oxygen into the engine allowing more fuel to be burnt and increasing performance. Now that's exactly what I'm going to outline for you in this presentation, how you can bolt on some currently available technology to increase the performance of your applications. 3. coherenceI'm going to start with an overview of the supercharger - Oracle Coherence - what it is and how it works. And then show how it canbe used and is being used by our customers to increase performance, reduce costs and improve user experience.What does coherence mean? Well, its defined as the state of cohering or sticking together, or the logical and orderly and consistent relation of parts. I like that, logic, order, consistency. 4. If you want a 3 second response time. Youve got 3 seconds. 1 in the presentation layer, 1 in the mid-tier and 1 in the back end This quote from the CIO of one of our customers really sums up the problem space that were in here. If you want a n second response time then you only have n seconds to deliver that. There are no silver medals for second place in the response time Olympics. 5. wrote the book 6. David Chappell Enterprise Service Bus 7. Avoid processing wherever possible What does a scalable architecture need to ensure? And see if you can identify a theme here... 8. Avoid I/O where possible 9. Avoid serialization and deserialization 10. Avoid sending a large document around 11. Avoid distributed file schemes 12. Avoid driving increased traffic to back end systems I like this one a lot... and it's a classic SOA example, and if it's a mainframe there's a monetary cost in terms of haw many MIPS you need to pay for, using the technologies described here you can actually reduce the traffic to the back ends, reduce the MIPS that you're paying for. 13. Value simplicity.Maximize the amount of work not done Thanks to the Agile Alliance for this one more details here 14. where are we today? 15. Thanks to the Lego block analogy to explain SOA interoperable, unbreakable, composable, reusable. 16. paradigm shift 17. business agility 18. flexibility 19. We're trying to achieve business agility, business alignment, so that when a line of business comes to IT and says government legislation on superannuation just changed, make our systems change to reflect it. IT can say sure thing rather than suck it's teeth and say, "it's gonna cost ya" like some dodgy plumber. 20. large payloads 21. unexpected usage 22. unmet SLAs 23. tearing down silos 24. hard to share 25. boundary costs 26. And then there's this organisational tension that developing between the SOA architects who come up with these pristine blueprints for how the IT world should look within their organisation and then they hand that off 27. to IT operations and say make that run, make it highly available, make it high performance, ensure that it will scale. And today there is too tight a coupling between those two sets of folks - between the people that design the architecture and the people that have to operate it. For example, what does it mean to scale up the order to cash process? That process might consist of 42 different services across 8 machines. How do IT ops go about bringing some scale to that process at the same as managing the end of month financial close and keeping the lights on and fixing the printer on level 6 while resetting my password? 28. Scalability 29. Reliability 30. Availability always open, always on. 24 hours a day, 7 days a week. 31. stateless 32. idempotent 33. So thinking about state and state management we have this continuum that runs from completely stateless to fully stateful. and there are a few things that can affect or influence this - one is complexity - the more complex a service the more likely it is that it will need to manage state across multiple requests coming in and out of the service, and the longevity of the service - how long the service needs to run to satisfy a particular business transaction. Service Sophistication, Longevity Complex Stateless Cookies + ServletSession State passing via XML Payloads Loose Coupling, Tight Coupling Service Complexity Stateful Service State Repository Simple 34. state aware So that's where we are... where should we be? 35. continuous availability 36. predictable linear scalability 37. Next pop quiz question, who knows what this is? Thats right, Napster. Remember Napster? Well think of Coherence like Napster for your business data - highly available, highly distributed peer to peer clustering protocol. 38. P B BackupNode Primary Node Put() HashKey/CacheKey Hashmap iFace Here's a picture of how it works. At the core are a set of connected network nodes - software processes, which cooperate together across a network, with the sole purpose of storing and maintaining instance data for individual application objects. An application does a put, that operation is delegated to the grid and the grid automatically elects a primary storage node and a backup storage node on another machine and it does that in a highly available way. So each of these nodes is on a different physical server, and each knows about each other, they are aware of each other existence, health, availability etc... If the primary owner goes down the grid detects that and selects a new primary and backup pair. And this switching this failover can happen in flight without affecting a transaction because of the self-healing nature of the grid. Application Object Application Object Application Object 39. Put() HashKey/CacheKey Hashmap iFace P B BackupNode Primary Node X Application Object Application Object Application Object 40. Put() HashKey/CacheKey Hashmap iFace P B BackupNode Primary Node X Application Object Application Object Application Object 41. BackupNode Primary Node Write Behind Queue DB Grid Put() HashKey/CacheKey ESB MediationP B Service Service Service 42. Shared web session state across multiple portals 43. Caching the results of service calls 44. State management for stateful services 45. Process large XML payloads 46. Shared web session state across multiple portals 48.

  • Solution
  • Coherence solution provides caching for the data and no loss of connections to the back-end systems
  • Customers can shop between different brands - Gap, Banana Republic, Old Navy and Piperlime without having to check out between sites
  • Problem
  • On-line store shopping cart would lose connections to back-end servers, causing users to re-select the items and add to shopping cart, causing loss of sales and low customer satisfaction
  • Slow on-line store performance
  • Scenario
  • Online retail store
  • High availability of the web site, shopping cart, over 4 distinct brands

49. Caching the results of service calls 51.

  • Solution
  • Coherence solution provides caching side-cache - for the multi page sales histories
  • One call to the back end and then 9 calls to the cache to get 10 pages of data.
  • Problem
  • Costly to continue to hit the back end for every next page of data
  • Slow on-line performance in the field
  • Un-necessary interaction
  • Scenario
  • Sales application
  • Large volumes of sales data retrieved multi page histories

52. State management for stateful services 54.

  • Solution
  • Moving event processing into application tier increased capacity to handle peak loads
  • Enabled application developers to modify logic without impacting the database; operational cost savings & increased flexibility
  • Problem
  • Matching engine supporting several thousand matches per second, with intense hot spots on specific instruments
  • Revenue tied directly to customer activity. Need for high-throughput, low-latency solution for financial transactions
  • Scenario
  • Expanding their infrastructure to handle increased traffic
  • Looking for an event-driven architecture, treating bids as incoming events, modifying the state of bidding markets, and dispatching matched bids

55. Process large XML payloads 56. So this picture is fairly typical - every vendor of a Service Bus could draw out this picture - they all use some sort of Message Oriented Middleware (MOM) to achieve high reliability and quality of service - and I'm using MOM here to cover anything that's JMS based or MQ based or Rendezvous based or implements WS-RM - all those similar techniques of message retries and store and forward to reliably send data from one place to the next... And then it dawned on us, why do I have to send something, when it really doesn't have to go anywhere? Everything is already in the grid. All interested parties can act on it. Why do I have to put it in a big JMS message or a bi