The Artima Developer Community
Sponsored Link

Make Room for JavaSpaces, Part VI
Build and Use Distributed Data Structures in Your JavaSpaces Programs
by Susan Hupfer
First Published in JavaWorld, October 2000

<<  Page 5 of 15  >>


A Distributed MP3 Encoding Application

Most folks have heard of MP3, a wildly popular sound compression format. Sounds encoded in MP3 format almost match CDs in quality, but MP3 files are much smaller than the equivalent files on a CD. If you've ever played around with digital audio files, you may have performed encoding -- the process of taking uncompressed digital audio data (for example, wav files on your PC) and compressing them according to a specific compression scheme such as MP3. MP3 encoding can be a computationally intensive process, especially if the original wav file is large or you have a large collection of wav files to encode.

You might imagine submitting your wav files to a high-powered, net-connected MP3 encoding service that would perform the wav-to-MP3 encoding for you and return the MP3 data. If many users simultaneously send encoding requests to a single MP3 encoder, the requests would need to queue up until they could be serviced. However, since the encoding tasks are completely independent of each other, they lend themselves well to being serviced in parallel by a farm of MP3 encoders. By using a space, you can easily build a distributed MP3-encoding application; Figure 3 shows the architecture of the system.

Figure 3. The architecture of the distributed MP3-encoding application

At first, you might think that your distributed application should just use bags, as the compute server did. In other words, the application could revolve around a bag of MP3 requests (deposited by MP3Requester processes and picked up by MP3Worker processes) and a bag of MP3 results (deposited by MP3Workers and picked up by MP3Requesters). This scheme has one major drawback: there is no guarantee of fairness. In other words, there's no assurance that some newer encoding requests won't be serviced before older requests. In fact, some requests might languish forever and never get picked up, while others are serviced. Ideally, you'd like for encoding requests to be handled on a first-come, first-served basis, with a guarantee of eventually being serviced (provided that at least one MP3 worker remains up and running to perform encoding). It turns out that a channel distributed data structure is exactly what you need to accomplish those goals.

As you can see from Figure 3, MP3Requester processes add MP3 encoding requests to the tail end of an MP3 Request channel, and MP3Worker processes take the requests off the head end of the channel, in order. An MP3Worker calls a piece of third-party software to perform the actual wav to MP3 encoding; when that software generates the MP3 data, the worker writes the data to a result bag in the space (you can assume for this example that the order in which results are picked up is not important). In the meantime, the MP3Requester processes constantly monitor the space to pick up and display results that have been tagged for them.

Now that you have the aerial view of the application, I'll dive down into the details.

<<  Page 5 of 15  >>

Sponsored Links

Copyright © 1996-2018 Artima, Inc. All Rights Reserved. - Privacy Policy - Terms of Use