Memory, in general, is not an issue here Sockets in RMI arent a limitation either
6.2.1.1 Memory, in general, is not an issue here
As long as the instances of servers can comfortably reside within a single JVM, memory is not an issue. The amount of memory required by a server can be divided into two pieces: client-specific state and general-purpose state. For example, in an e-business application, the clients current order is clearly client-specific; the general catalog of items available for sale is general-purpose. Generally speaking, the amount of memory required by a set of servers is a constant the general-purpose state plus an amount proportional to the number of currently active clients. However, the latter value is often proportional to the number of currently open socket connections. This is one example of why memory is not going to be the bottleneck. Youll swamp other scarce resources long before memory becomes a problem that cant be solved by popping more memory modules in the machine. This might seem false in the case of our accounts. If there are 25 million accounts, then the amount of memory required for 25 million server objects would seem to be substantial, independent of the number of clients currently connected. The factory pattern, which we discuss in Chapt er 17 , is designed to explicitly handle this problem.6.2.1.2 Sockets in RMI arent a limitation either
It used to be that socket allocation was a major resource limitation. This was caused by the combination of two factors: • Processes on most major operating systems were only allowed to have a limited number of sockets open. [ 3] In fact, the actual limit is usually the number of file descriptors a process can have open; file descriptors are used for both files and sockets. [ 3] The limit is built into the operating system and is usually less than or equal to 1024. • Each server required an instance of ServerSocket to listen for connections. The combination of these two factors is deadly. It means that a very small number of servers can be running inside a single process. And since launching a process in Java requires launching a JVM, which consumes a significant amount of memory and operating-system resources, this can quickly become a major issue. RMI [ 4] solves this problem by reusing sockets. In other words, if a client JVM sets up a socket connection to a server JVM, then the connection is actually kept alive for a short period of time by the RMI infrastructure. If, after the client request has been handled, a second request is made from the same client JVM, that request will reuse the same socket connection. This means that the number of socket connections required by an RMI server is approximately: 1 + number of simultaneous requests. [ 4] That is, the Sun Microsystems, Inc. implementation of RMI. Socket sharing isnt actually required by the RMI specification. This is only an approximation because there may be temporary open, unused socket connections that correspond to completed requests. In practice, this can be significant. If you find that unused sockets are being retained for long periods of time and constitute a significant resource limitation, then you can either set parameter values to configure how long the RMI runtime keeps unused sockets open well discuss this, and other settings, in Chapt er 16 or use a custom socket factory to achieve a similar effect well discuss custom socket factories in Chapt er 18 . Note that this number is entirely dependent on the number of clients and how busy they are. It does not depend at all on the number of servers. Socket reuse is actually a fairly significant benefit to using RMI. Its not all that hard to implement, but doing it right requires a fair amount of code and some forethought. The RMI Runtime Now that weve discussed sockets, its time to admit that our architectural diagrams are hiding a bit of the complexity of RMI. Neither stubs nor skeletons use sockets directly. Instead, stubs use an object called a RemoteRef , which handles the actual details of communicating with a remote process. This extra layer of indirection allows the RMI infrastructure to manage network communications and conserve scarce resources. From the networking point of view, it makes both socket sharing and distributed garbage collection possible. It also enables RMI to effectively share a thread pool across multiple servers. Well explore this much more fully in Par t I I . The networking details are covered in Chapt er 16 , and thread pools are discussed in Chapt er 11 and Chapt er 12 . Well ignore the RMI runtime for the rest of this chapter and for the rest of Part 1. However, its not a bad idea to keep the fact that sockets are shared in the back of your mind. Its also good to remember that when we draw a stub and a skeleton in a picture, we really mean something like that shown in Figur e 6- 1 . Figure 6-1. Stubs and skeletons interact with the RMI runtime and not directly with the network6.2.1.3 An example of a resource limitation
Parts
» OReilly.Java.Rmi. 2313KB Mar 29 2010 05:03:49 AM
» Writing data Resource management
» Some Useful Intermediate Streams
» Revisiting the ViewFile Application
» Protocols Metadata Protocols and Metadata
» The accept method A Simple Web Server
» Customizing Socket Behavior Sockets
» Direct Stream Manipulation Subclassing Socket Is a Better Solution
» A Special-Purpose Socket Special-Purpose Sockets
» Factories Socket Factories Special-Purpose Sockets
» Registering providers Using SSL with JSSE
» Configuring SSLServerSocket Using SSL with JSSE
» A Network-Based Printer A Socket-Based Printer Server
» The Basic Objects A Socket-Based Printer Server
» DocumentDescription Encapsulation and Sending Objects
» ClientNetworkWrapper Network-Aware Wrapper Objects
» ServerNetworkWrapper Network-Aware Wrapper Objects
» Passing by Value Versus Passing by Reference
» The Architecture Diagram Revisited
» The Printer Interface Implementing the Basic Objects
» Examining the skeleton Implementing a Printer
» DocumentDescription The Data Objects
» The Client Application Summary
» The Bank Example Introducing the Bank Example
» Security Scalability Design Postponements
» The Basic Use Case A Distributed Architecturefor the Bank Example
» Partial Failures Problems That Arise in Distributed Applications
» Network Latency Problems That Arise in Distributed Applications
» Memory, in general, is not an issue here Sockets in RMI arent a limitation either
» Applying this to Bank versus Accounts
» Should We Implement Bank or Account?
» Iterators, again Applying this to the Account interface
» Applying this to the Account interface
» Data Objects Dont Usually Have Functional Methods Interfaces Give You the Data Objects
» Accounting for Partial Failure
» A Server That Extends UnicastRemoteObject A Server That Does Not Extend UnicastRemoteObject
» The benefits of UnicastRemoteObject
» The costs of UnicastRemoteObject
» Getting Rid of the Skeletons
» Build Test Applications The Rest of the Application
» Dont Hold Connections to a Server Youre Not Using
» Validate Arguments on the Client Side Whenever Reasonable
» The Actual Client Application
» Deploying the Application The Rest of the Application
» Drilling Down on Object Creation
» The write methods ObjectOutputStream
» The stream manipulation methods Methods that customize the serialization mechanism
» The read methods ObjectInputStream
» Declaring transient fields Implementing writeObject and readObject
» Implement the Serializable Interface Make Sure That Superclass State Is Handled Correctly
» The Data Format The Serialization Algorithm
» Writing A Simplified Version of the Serialization Algorithm
» annotateClass replaceObject RMI Customizes the Serialization Algorithm
» Maintaining Direct Connections The Serialization Algorithm
» The Two Types of Versioning Problems
» How Serialization Detects When a Class Has Changed Implementing Your Own Versioning Scheme
» Serialization Depends on Reflection Serialization Has a Verbose Data Format
» It Is Easy to Send More Data Than Is Required
» Comparing Externalizable to Serializable
» The Calling Stack Basic Terminology
» The Heap Threads Basic Terminology
» Mutexes Applying This to the Printer Server
» Controlling Individual Threads Threading Concepts
» Coordinating Thread Activities Threading Concepts
» Cache Management Assigning Priorities to Threads
» The effects of synchronization on the threads local cache
» The wait methods The notify methods
» Starting a thread is easy Stopping a thread is harder
» Using Runnable instead of subclassing Thread Useful methods defined on the Thread class
» The Basic Task Implementing Threading
» Applying this to the bank example
» Synchronize around the smallest possible block of code
» Dont synchronize across device accesses
» Concurrent modification exceptions Be Careful When Using Container Classes
» Start with Code That Works Use Containers to Mediate Interthread Communication
» Immutable Objects Are Automatically Threadsafe Always Have a Safe Way to Stop Your Threads
» Pay Careful Attention to What You Serialize
» Use Threading to Reduce Response-Time Variance Limit the Number of Objects a Thread Touches
» Acquire Locks in a Fixed Order Use Worker Threads to Prevent Deadlocks
» The Idea of a Pool Two Interfaces That Define a Pool
» A First Implementation of Pooling
» Problems with SimplePool Pools: An Extended Example
» The Creation Thread Pools: An Extended Example
» Gradually Shrinking the Pool
» What Were Testing Testing the Bank Application
» When Are Naming Services Appropriate?
» bind , rebind , and unbind lookup and list
» Bootstrapping the Registry The RMI Registry Is an RMI Server
» Querying the Registry Launching an Application-Specific Registry
» Filesystems Yellow pages The general idea of directories and entries
» Security Issues The RMI Registry
» Operations on contexts Hierarchies
» Attributes are string-valued, name-value pairs
» Federation Federation and Threading
» Value Objects Represent Sets and Lists Paths, Names, and Attributes Are All Distinct
» AttributeSet The Value Objects
» Path and ContextList The Value Objects
» The Context Interface The Java Naming and Directory Interface JNDI
» Using JNDI with the Bank Example
» How RMI Solves the Bootstrapping Problem
» Ordinary Garbage Collection Distributed Garbage Collection
» Defining Network Garbage Distributed Garbage Collection
» Leasing Distributed Garbage Collection
» The Actual Distributed Garbage Collector The Unreferenced Interface
» The Standard Log RMIs Logging Facilities
» The Specialized Logs RMIs Logging Facilities
» java.rmi.server.randomIDs sun.rmi.server.exceptionTrace
» sun.rmi.dgc.client.gcInterval sun.rmi.dgc.server.gcInterval
» sun.rmi.dgc.checkInterval sun.rmi.dgc.cleanInterval
» Resource Management Factories and the Activation Framework
» A Basic Factory Implementing a Generic Factory
» The new factory Building on the Account-Locking Mechanism
» The new account The launch code and the client
» Persistence and the Server Lifecycle
» Making a server into an activatable object
» Deploying an Activatable System
» ActivationDesc, ActivationGroupDesc, and ActivationGroup in More Detail
» Shutting Down an Activatable Server
» -port -log rmid Command-Line Arguments
» sun.rmi.server.activation.debugExec
» A Final Word About Factories
» Implementing Serializable Implementing equals and hashCode
» Modifying Ordinary Servers Incorporating a Custom Socket into an Application
» Modifying Activatable Servers Incorporating a Custom Socket into an Application
» Interaction with Parameters Incorporating a Custom Socket into an Application
» A Redeployment Scenario How Dynamic Classloading Works
» A Multiple-Deployment Scenario How Dynamic Classloading Works
» Requesting a Class The Class Server
» Receiving a Class Handling JAR files
» Suns Class Server The Class Server
» Server-Side Changes Using Dynamic Classloadingin an Application
» Naming-Service Changes Using Dynamic Classloadingin an Application
» Client-Side Changes Disabling Dynamic Classloading Entirely
» A Different Kind of Security Problem
» AWT permissions The Types of Permissions
» File permissions Socket permissions
» Property permissions The Types of Permissions
» Installing an Instance of SecurityManager
» How a Security Manager Works java.security.debug
» Using Security Policies with RMI Policy Tool
» Printer-Type Methods Report-Type Methods
» Client-side polling Polling code in the printer application
» Server-side callbacks Define a client-side callback interface
» Implement the client-side interface
» Server-evaluation models Ch a pt e r 7
» Iterators on the client side
» Implementing Background Downloading on the Client Side
» The Common Gateway Interface Servlets
» Naming services and the server machine
» The Servlet Code A Servlet Implementationof HTTP Tunneling
» Modifying the Tunneling Mechanism
» Disabling HTTP Tunneling HTTP Tunneling
» Defining the Interface Generating Stubs and Skeletons
» The Server The Launch and Client Code
Show more