Appendix C. JavaSpacesJavaSpaces is a new distributed object system being proposed by Sun as a package at a higher level than the existing RMI and object serialization facilities built into Java. JavaSpaces provides a distributed, persistent object system that is roughly modeled after earlier shared memory systems, such as LINDA. While it has some analogies with parallel shared memory systems such as the Posix shm_xxx library and shared memory facilities in parallel languages like Python, it also has some important differences. This appendix provides an overview of the general JavaSpace architecture as currently described in the draft JavaSpace specification. It doesn't go into detail about how to use the JavaSpace API, since the API isn't fully available yet. This appendix only includes the core elements of the specification, without discussing any proposed features that may or may not be in the final API. C.1. Overview of JavaSpacesThe distributed application paradigm supported by JavaSpaces is one in which remote agents interact with each other indirectly through shared data object spaces. Objects are stored in a JavaSpace in the form of entries. Clients write entries into the space, read entries from the space, or take entries from the space, as shown in Figure C-1. Figure C-1. JavaSpaces general architectureAccess to the entries in JavaSpaces is through a small set of basic operations:
Multiple basic operations can be assembled into transactions that group basic operations into a single, atomic aggregate operation. There can be many clients and many JavaSpaces in a given distributed application. One client, and even one transaction from one client, can access multiple JavaSpaces. So instead of one agent sending a message to another, or invoking a remote method directly on another object within an agent, agents interact by writing and reading objects in JavaSpaces. An important feature of the JavaSpaces specification is that all operations on a given JavaSpace are considered unordered. If you have multiple threads or multiple remote agents issuing operations on a JavaSpace, and for some reason you want to impose some order on the operations, then it's up to you to synchronize your threads or agents as needed. Each JavaSpace holds data in the form of entries, which can either be read, written, or taken from a JavaSpace. Each entry has one or more fields that are used to match incoming requests from clients. Each request to read, take, or be notified about an entry includes a template for the entry to match. In order for an entry in the JavaSpace to match, the entry must be of the same type as the template object. Each field in the template can either have a non-null value, which must match the fields in a matching entry in the JavaSpace, or a null value, which matches any value in that field. All operations on JavaSpaces are "transactionally secure," which means that each operation or transaction is either entirely committed or entirely noncommitted to a JavaSpace. So if a write to a JavaSpace succeeds, then you can be assured that the Entry was written and will appear in the next client read or take operation on the space. An operation on JavaSpaces can be either in the form of a simple operation, or a group of operations within a single Transaction. The authors of the JavaSpace specification make a point of distinguishing JavaSpaces from a distributed database system. A JavaSpace knows the type of its entries, and can compare field values, but it doesn't understand anything about the structure of the data in its entries. It also isn't meant to provide opaque read/write access to persistent data. An entry in a JavaSpace is a serialized copy of the object written to the space, and entries returned to clients as a result of read or take operations are separate copies of the objects in the space. Copyright © 2001 O'Reilly & Associates. All rights reserved. |
|