Orleans Cloud Computing
Download as PPTX, PDF3 likes882 views
The document provides an overview of the Orleans cloud computing framework, detailing its actor-based model and focus on enabling large-scale application development without in-depth knowledge of distributed systems. It discusses key concepts such as grains, asynchronous message passing, promises, persistence, and automatic scalability, along with the Orleans runtime that manages these aspects. Additionally, it highlights applications like Chirper and a linear algebra library, underlining the framework's capabilities in handling high workloads and its relevance in contemporary cloud software development.
Orleans Cloud Computing
- 1. Orleans Cloud Computing Huzaifa Salman Zafar
- 2. Intro to Orleans cloud computing Cloud computing Diverse clients(pc’s, smartphones, sensors etc.) Computation and data storage in cloud (cloud=internet) Cloud systems are distributed systems Distributed systems A distributed system is a software system in which components located on networked computers communicate and coordinate their actions by passing messages
- 3. Message Passing A type of communication between process Messages may include functions, signals, and data packets Works as follows A message is sent to a process Process may be an actor
- 4. Actor model The Actor model was introduced in 1973 by Hewitt, Bishop and Steiger. Mathematical model of concurrent computation Actor serves as basic building block of concurrent computation. In response to a message received by actor, actor can Make local decisions Create more actors Send more messages etc
- 5. Grains Basic programming unit All the code run within grains Isolated units of state and computation Communicate through asynchronous messages Messages are method calls asynchronous messages Message is sent and does not need to wait for reply Opposite of synchronous message passing
- 6. Promises Within a grain, promises are mechanism for Managing asynchronous message passing Local task based concurrency Concurrent computation Several threads of operation may execute concurrently
- 7. Promises |continued…. Communication between grains Asynchronous message passing Orleans messages are method calls Message call returns immediately with a promise for a future result An application can bind code to the promise which will execute when result arrives it can treat the promise like a future and explicitly wait for a result.
- 8. Goals of Orleans Enable the developers unfamiliar with distributed systems to build large scale applications That these systems handle multiple orders of large magnitude of growth without requiring redesign or rearchitecture.
- 9. Activations To handle increase load on a grain And to provide higher system throughput Orleans automatically creates multiple instantiations called activations The activations process independent requests for the grain, possibly across multiple servers These activations cannot share memory or invoke each other’s methods
- 10. Persistence A cloud system has persistent state that is kept in durable storage. Orleans integrates persistence into grains Persistent grain state means State Changes in one activation of a grain will be available for subsequent activations of the grain Multiple activations can modify grain’s state Orleans provides a mechanism to reconcile changes
- 11. Automated Scalability A service is scalable if increased load can be handled by proportionally increasing server capacity Techniques used for achieving scalability in orleans Asynchrony and replication
- 12. Programming model Promises Grains execution model Error handling Grain interfaces Grain references Creating and using grains Grain classes
- 13. Promises Orleans uses promises as its asynchrony communication Promises have life cycle Unresolved (expecting a result in unspecified time) fulfilled (result is received, value of promise=result) Resolved(fulfilled or broken is considered resolved. ). Implemented as .NET types
- 14. Grain Execution Model When an activation receives a request, it processes it in discrete units of work called turns All grain code execution, runs as a turn Many turns belonging to different activations may run in parallel But each activation executes its turns sequentially Hence, execution in an activation is logically single threaded.
- 15. Grain interfaces Orleans uses standard .NET interfaces to define the interface to grain’s services Grain References A grain reference is a proxy object that provides access to a grain. A proxy object acts as an intermediary between the client and an accessible object. The purpose of the proxy object is to forward calls to the accessible object only if it is not destroyed.
- 16. Creating and Using Grains For each grain interface, Orleans generates a static factory class and an internal proxy class Clients use the factory classes to create, find, and delete grains. The proxy classes are used by the Orleans runtime to convert method calls into messages.
- 17. Orleans runtime The mechanisms that Orleans provides for an application is called Orleans runtime Orleans is a framework for .NET runtime that can be used from any .NET language (C#, F#, etc. Orleans can run on desktop machines, servers running Windows Server 2008, and the Microsoft Windows Azure cloud platform.
- 18. .NET Framework Is a software framework developed by Microsoft that runs primarily on Microsoft windows The easiest way to build apps on the Microsoft platform Provides Large class library (Framework class library) A software environment known as common language runtime (CLR) CLR provides security, memory management, and exception handling etc
- 19. Orleans runtime (continued) The state of a grain is managed by the Orleans runtime initialization, replication, reconciliation, and persistence The programmer identifies the persistent state and Orleans handles the rest
- 20. Orleans runtime (continued) Persistence Activate method, Handle requests (operate in memory) Wait for completion of request , call deactivate Write to persistent storage Replication An activation is single threaded, so processing is limited But Orleans uses grain replication –multiple activations of a grain And achieve elasticity and scalability
- 21. Orleans runtime (continued) Isolation Activations of many different grains, as well as multiple activations of the same grain are isolated all activations communicate only through asynchronous message passing and reference each other using grain references (proxy objects). This allows Orleans to place activations on any server, even across data centers, and migrate activations between servers, in order to balance load
- 22. Orleans runtime (continued) Resource Management growing and shrinking the number of activations of a grain Requests are initially randomly distributed to existing activations A request arriving at an overloaded server is rejected
- 23. Orleans runtime (continued) State Reconciliation If multiple activations of a grain concurrently modify their persistent state, the changes must be reconciled into a single, consistent state. a last-writer-wins strategy data structures (records, lists, and dictionaries) that track updates and automatically reconcile conflicting changes
- 24. Orleans runtime (continued) Transactions A transaction is created at the arrival of an initial, external request from a client A transaction is completed when the request processing finishes execution. It is committed when its changes are written to durable storage
- 25. APPLICATIONS Here we discuss 3 applications Chirper large-scale Twitter-like publish-subscribe system for distributing small text message updates within a large network of consumers / subscribers Linear Algebra Library General purpose computations on scalars, vectors, and matrices (including higher dimensions as tensors).
- 26. Chirper A Chirper account is naturally modeled as a grain The account grain exposes three public interfaces: IChirperAccount, IChirperSubscriber and IChirperPublisher IChirperAccount single user account, start and stop following another user IChirperSubscriber and IChirperPublisher are used for subscription and notification activities An account that receives or sends large number of messages is replicated to balance load
- 27. Linear Algebra Library The core of a linear algebra library is the vector-matrix multiplication This operation is the basis for many algorithms, including PageRank, clustering, vector-matrix multiply is quite simple if data is held in memory on one machine Consider a web graph greater than 10^11 pages with more than 10^12 links The Web graph is the graph of the Web pages together with the hypertext links between them.
- 28. Linear Algebra Library The computations are broken into worker grains that own pieces of the data set. Special coordinator grains manage the computation by dynamically assigning work to worker grains
- 29. Large Graph Engine Graphs are central to web search, social networking, and other web applications Large graphs pose many challenges they do not fit a single computer and distributed algorithms are communications intensive partitioning and distributing graph data (nodes, edges, and metadata) across many machines represent a partition of the nodes by a grain
- 30. Large Graph Engine Every server hosts a small number of partition grains, and every partition grain contains a moderate number of graph data nodes (10^4 – 10^6) A graph algorithm running in a partition directly accesses nodes in its partition Accesses across partitions involve sending messages between partition grains it imposes no restrictions on the size of a grain . Grains can hold potentially large amounts of state
- 31. PERFORMANCE MEASUREMENTS Chirper We created a synthetic network of 1,000 user accounts each user following 27 random users ran load generator processes on multiple machines each generator posting messages of varying size The system can deliver approximately 7,000 chirps per second (1600 tweets per second by
- 32. Related Work Actors Actors are a well-known model for concurrent programming Orleans extends the basic actor model with support for replication, transactions, and consistency E is an object-oriented programming language for secure distributed computing. E has a concurrency model similar to Orleans
- 33. Related Work Distributed Object Models Enterprise Java Beans (EJB), Microsoft’s Component Object Model (COM), and the Common Object Request Broker Architecture (CORBA) Based on distributed objects primarily synchronous RPC They share Orleans’s goals of offering a higher-level collection of abstractions that hide some of the complexity of building distributed systems
- 34. FUTURE WORK An on-going area of research is resource management when and where should a new activation be created, rather than reusing an existing one? When should an existing activation be deactivated? important area for future work is in extending Orleans to run on devices such as PCs and smartphones
- 35. CONCLUSIONS This paper described the design and implementation of Orleans actor-like model of isolated, replicated grains that communicate through asynchronous messages and manage asynchronous computations with promises We believe that the Orleans framework can significantly simplify the development of cloud applications This is not a strong guarantee, as it is possible to write a bad program in any language
- 36. CONCLUSIONS Cloud services achieve high throughput by processing multiple, independent requests concurrently Orleans supports a simple, single threaded model within a grain, but permits parallelism between grains, although limited to message passing Computers and networks fail in distributed systems, so error handling and recovery code is fundamental.
- 37. CONCLUSIONS Cloud applications must respond to varying and unpredictable workloads. Grain replication offers a simple, mostly transparent mechanism that permits Orleans to allocate more computing resources at bottlenecks in an application Orleans is currently being used by several projects inside Microsoft Research that training and education remains an important aspect of cloud software development
- 38. Thank You
Editor's Notes
- #12: Scalability is the ability of a system, network, or process to handle a growing amount of work in a capable manner .
- #17: CreateGrain immediately returns a grain reference , This enables pipelining of asynchronous requests to the grain, such as FollowUser
- #19: FCL provides user interface, data access, database connectivity, numeric algorithms etc.
- #21: The Orleans runtime activates a grain with its persistent properties already initialized The grain’s Activate method is then called to allow it to initialize its transient state The runtime then invokes methods to handle requests sent to the activation which can operate freely upon the state in memory. To commit an activation to storage, the runtime waits for the completion of a transaction (i.e., the end of a request), calls the grain’s Deactivate method, and writes the grain’s state property values to persistent storage
- #28: PageRank is an algorithm used by Google Search to rank websites in their search engine results