Check out the new USENIX Web site. next up previous
Next: On-the-fly compression Up: Existing mechanisms for network Previous: Downloading individual files

Downloading Jar archives

Another standard network class loading mechanism in Java is to download Jar archive files [10]. Jar archives use the same underlying format as Zip files [5]. Each file in the archive is usually (though not always) compressed using the Deflate algorithm [4].

Jar files are convenient for application developers because they provide a simple mechanism for packaging an entire application or a component of an application into a single file. Since Jar files contain multiple files, they can be considered a form of pre-sending by the server, meaning that the network latency cost of the request is paid only once, for the entire Jar file, rather than once for each class or resource file. Jar files also offer random access to the files contained in the archive, which is useful for loading classes and resources from a Jar archive on disk. However, when used to transport classes and resources over a network they are likely to deliver unneeded files or to deliver files in the wrong order, because they typically contain a large number of files. In addition, the fact that the files are compressed individually results in a lower compression ratio than would be possible if the entire archive were compressed using cumulative compression.

JDK version 1.3 introduced Jar indexes. A Jar index is a mapping from class and resource files to the names of the Jar files in which they are located. The Jar index for a set of Jar files is contained in a master Jar file. Once the master Jar file and the Jar index are received by the JVM, subsequent Jar files are downloaded only if they contain a class or resource file needed by the running application. This feature is an improvement over the technique used by earlier JDK versions to find class and resource files in a collection of Jar files: each Jar file was simply downloaded in sequence until the desired file was located. Through the use of a Jar index, an application can be split up into discrete components which are downloaded only when needed. This can help reduce the number of unneeded classes and resources downloaded. (The functionality of the Jar index is in some ways similar to our concept of bundling, which will be described in section 2.)


next up previous
Next: On-the-fly compression Up: Existing mechanisms for network Previous: Downloading individual files

David Hovemeyer
Tue Feb 27 18:43:09 EST 2001