Distributed Object-based systems Prof. Orhan Gemikonakli Module Leader: Prof. Leonardo Mostarda Università di Camerino Prof. Orhan Gemikonakli - Camerino,

Prof. Orhan Gemikonakli - Camerino, Last lecture Basic concepts Fault modelling Failure masking RPC Semantics in the Presence of Failures Reliable multicast schemes Virtual Synchrony Two phase commit |Three phase commit Checkpointing and Message logging Prof. Orhan Gemikonakli - Camerino,

Prof. Orhan Gemikonakli - Camerino, Outline Architecture Processes Communication Naming Synchronization Consistency and Replication Fault Tolerance Tanenbaum & Van Steen, Distributed Systems: Principles and Paradigms, 2e, (c) 2007 Prentice-Hall, Inc. 0-13-239227-5 Prof. Orhan Gemikonakli - Camerino,

Prof. Orhan Gemikonakli - Camerino, Learning outcomes To understand the basic concepts related to distributed object-based systems To describe and discuss Architecture Processes Communication Naming Synchronization Consistency and Replication Fault Tolerance as they apply to distributed object-based systems Prof. Orhan Gemikonakli - Camerino,

Distributed Object-based systems A paradigm used to organise ditributed systems The notion of object plays a key role in establishing distribution transparency Everything is treated as objects Clients are offered services and resources in the form of objects they can revoke. Examples of object-based distributed systems CORBA Java-based systems Globe Prof. Orhan Gemikonakli - Camerino,

Prof. Orhan Gemikonakli - Camerino, Architecture Object orientation plays an important part in software development Functionality can be added as independent components 1980s: Distributed Object-oriented systems started An independent object hosted by a remote server High degree of distribution transparency An object Encapsulates data (state) Encapsulates operations on that data (methods) May implement multiple interfaces Prof. Orhan Gemikonakli - Camerino,

Prof. Orhan Gemikonakli - Camerino, Distributed Objects Distributed object: an interface at one machine while the object resides on another machine When a client binds to a distributed object an implementation of the object’s interface called proxy is then loaded into the client’s address space. A proxy marshals method invocations and unmarshalls reply messages (similar to a stub in RPC) Skeleton: The server-side stub Objects: compile-time objects and run-time objects Prof. Orhan Gemikonakli - Camerino,

Prof. Orhan Gemikonakli - Camerino, Distributed Objects Figure 10-1. Common organization of a remote object with client-side proxy. Prof. Orhan Gemikonakli - Camerino,

Example: Enterprise Java Beans (EJBs)) Services: messaging, naming DB access etc. Figure 10-2. General architecture of an EJB server. Prof. Orhan Gemikonakli - Camerino,

Prof. Orhan Gemikonakli - Camerino, Four Types of EJBs The use of services is automated, but a programmer should make a distinction between types of EJBs: Stateless session beans Stateful session beans Entity beans Message-driven beans Prof. Orhan Gemikonakli - Camerino,

Prof. Orhan Gemikonakli - Camerino, Four Types of EJBs Stateless session bean: a transient object; invoked once, does it work, gets discarded (e.g. an SQL query) Stateful session bean: a persistent object; maintains client related state (e.g. electronic shopping cart). Yet, limited lifetime. Entity bean: a long time persistent object (e.g. holding customer details in e-commerce) Message-driven beans are used to program objects that should react incoming messages and send messages. Stateless. Prof. Orhan Gemikonakli - Camerino,

Example: Globe Distributed Shared Objects (1) Globe: an object-based distributed system Scalability plays a central role Can support large-scale wide-area systems and huge numbers of users. Objects (distributed shared objects) encapsulate state and operations on that state. Also encapsulate the policies on the distribution of an object’s state over its replicas Physically distributed and replicated (the states) (no remote object model) Have local implementations (with or without states) – only methods are visible Primitive or composite local object Prof. Orhan Gemikonakli - Camerino,

Globe Distributed Shared Objects (2) Figure 10-3. The organization of a Globe distributed shared object. Prof. Orhan Gemikonakli - Camerino,

Globe Distributed Shared Objects (3) Figure 10-4. The general organization of a local object for distributed shared objects in Globe. Prof. Orhan Gemikonakli - Camerino,

Prof. Orhan Gemikonakli - Camerino, Processes Object servers play a key role in object-based distributed systems. They are designed to host distributed objects Services are implemented by objects and are easy to change In multi-threaded object servers, each object can be assigned a thread, or each invocation can be assigned a thread. Prof. Orhan Gemikonakli - Camerino,

Processes: Object Adapter Activation: decisions on how to invoke an object Servant: piece of code that forms the implementation of an object (e.g. a java bean is a servant.) Object Adapter (Object Wrapper): A mechanism to group objects per policy. Controls one or more objects They are generic but configurable Several object adapters can exist in a server Invocation requests are first dispatched to the appropriate object adapters. Prof. Orhan Gemikonakli - Camerino,

Prof. Orhan Gemikonakli - Camerino, Object Adapter Figure 10-5. Organization of an object server supporting different activation policies. Prof. Orhan Gemikonakli - Camerino,

Example: The Ice Runtime System Ice is a distributed object system An object server in Ice is a process that starts by initialising Ice runtime system (RTS) Communicator: a runtime component that manages resources (e.g. pool of threads) Can specify maximum message length, maximum number of retries etc. Can create an object adapter Multiple communicators possible for an object server. E.g. managing different thread pools. Prof. Orhan Gemikonakli - Camerino,

Prof. Orhan Gemikonakli - Camerino, The Ice Runtime System Figure 10-6. Example of creating an object server in Ice. Prof. Orhan Gemikonakli - Camerino,

Prof. Orhan Gemikonakli - Camerino, Communication The mechanism of a client invoking a remote object is largely based on RPCs. Binding a client to an object Object references can be passed to processes and used for method invocations A process holding an object reference first binds to the referenced object. Binding results in a proxy been placed into the clients address space Implicit binding: provides the client with a simple mechanism to invoke methods directly Explicit binding: client first needs to call a function before binding. Prof. Orhan Gemikonakli - Camerino,

Binding a Client to an Object Figure 10-7. (a) An example with implicit binding using only global references. (b) An example with explicit binding using global and local references. Prof. Orhan Gemikonakli - Camerino,

Prof. Orhan Gemikonakli - Camerino, Parameter passing When invoking a method, objects may be passed with invocation Local objects (usually small) are passed as a whole Remote objects are passed by reference Prof. Orhan Gemikonakli - Camerino,

Prof. Orhan Gemikonakli - Camerino, Parameter Passing Figure 10-8. The situation when passing an object by reference or by value. Prof. Orhan Gemikonakli - Camerino,

Object-Based Messaging Remote Method Invocation (RMI) is the preferred way of handling communication in object-based distributed systems CORBA messaging combines method invocation and message-oriented communication Task: Compare and contrast RMI and CORBA in the context of handling communication in object-based distributed systems Prof. Orhan Gemikonakli - Camerino,

Prof. Orhan Gemikonakli - Camerino, Naming The interesting aspect of naming in object-based distributed systems evolves around the way that object references are supported. As discussed, in Java, these object references correspond to portable proxy implementations. Task: How are objects referenced in CORBA? Distinguish between referencing local, and remote objects. Prof. Orhan Gemikonakli - Camerino,

Prof. Orhan Gemikonakli - Camerino, Synchronisation Problem: Implementation details are hidden behind interfaces When a process invokes a remote object, it has no knowledge whther that invocation will lead to invoking other objects. If an object is protected against concurrent access, we may have cascading set of locks the invoking process is unaware of (Fig 10-a) When dealing with data resources such as files or database tables that are protected by locks, the pattern for the control flow is visible to the process. (Fig 10-b) Prof. Orhan Gemikonakli - Camerino,

Prof. Orhan Gemikonakli - Camerino, Synchronization Figure 10-14. Differences in control flow for locking objects Prof. Orhan Gemikonakli - Camerino,

Consistency and Replication General approach to replicated objects: Traditional approaches (as discussed in Chap 7) treating them as containers of data with their own operations Consistency: Data-centric consistency comes in the form of entry consistency No multiple invocations on the same object allowed Ensure that all the changes to a replicated object are the same. Prof. Orhan Gemikonakli - Camerino,

Consistency and Replication No multiple invocations on the same object allowed Use locks to allow only one method execution at a time – use local locks Recall that data is accessed through methods only, hence access to data is serialized Ensure thatr all the changes to a replicated object are the same. No two independent method invocations takes place on different replicas at the same time – order invocations so that each replica see them in same order. Use Primary-based approach or Totally ordered approach Prof. Orhan Gemikonakli - Camerino,

Prof. Orhan Gemikonakli - Camerino, Entry Consistency Figure 10-15. Deterministic thread scheduling for replicated object servers. Prof. Orhan Gemikonakli - Camerino,

Replication Frameworks (1) Invocations to objects are intercepted at three different points: At the client side just before the invocation is passed to the stub – in case the caller is replicated Inside the client’s stub, where the interception forms part of the replication algorithm – to forward or abort At the server side, just before the object is about to be invoked. Prof. Orhan Gemikonakli - Camerino,

Replication Frameworks (2) Figure 10-16. A general framework for separating replication algorithms from objects in an EJB environment. Prof. Orhan Gemikonakli - Camerino,

Replicated Invocations (1) Figure 10-17. The problem of replicated method invocations. Prof. Orhan Gemikonakli - Camerino,

Replicated Invocations (2) Figure 10-18. (a) Forwarding an invocation request from a replicated object to another replicated object. Prof. Orhan Gemikonakli - Camerino,

Replicated Invocations (3) Figure 10-18. (b) Returning a reply from one replicated object to another. Prof. Orhan Gemikonakli - Camerino,

Prof. Orhan Gemikonakli - Camerino, Fault Tolerance Similar mechanisms used for other distributed systems are used for object-based DSs too. CORBA: Replicates objects into object groups Object groups are referenced as a single object offering the same interface as replicas it contains Different replication strategies are used Primary-backup replication Active replication Quorum-based replication Object groups are referenced using an Interoperable Object Group Reference (IOGR) Prof. Orhan Gemikonakli - Camerino,

Example: Fault-Tolerant CORBA Figure 10-19. A possible organization of an IOGR for an object group having a primary and backups. Prof. Orhan Gemikonakli - Camerino,

An Example Architecture Figure 10-20. An example architecture of a fault-tolerant CORBA system. Prof. Orhan Gemikonakli - Camerino,

Example: Fault-Tolerant Java Causes for nondeterministic behavior: JVM can execute native code, that is, code that is external to the JVM and provided to the latter through an interface. Input data may be subject to nondeterminism. In the presence of failures, different JVMs will produce different output revealing that the machines have been replicated. Prof. Orhan Gemikonakli - Camerino,

Prof. Orhan Gemikonakli - Camerino, Summary Distributed Object-based Systems Architecture Processes Communication Naming Synchronization Consistency and Replication Fault Tolerance Prof. Orhan Gemikonakli - Camerino,

Prof. Orhan Gemikonakli - Camerino, Next Lecture Distributed File Systems Architecture Processes Communication Naming Synchronization Consistency and Replication Fault Tolerance Prof. Orhan Gemikonakli - Camerino,