MIME-Version: 1.0
Server: CERN/3.0
Date: Sunday, 01-Dec-96 19:20:04 GMT
Content-Type: text/html
Content-Length: 4447
Last-Modified: Friday, 05-Apr-96 16:35:05 GMT
Parallel MPEG Encoding
Parallel MPEG Encoding....... by MULVA Inc.
...And just what is this madness???
In the beginning in a lab deep in the heart of Upson Hall, there were
video tapes of classroom lectures. These tapes were on the shelf
collecting dust bunnies. What were we to do about this? Well, Brian
Smith decided that he wanted his lectures put online for his students
to view and thus began using RIVL
to compress his videos. He soon realized that compressing every
lecture would take an inordinate amount of time. He calculated that
if we could use every machine in the computer science department, it
would not take nearly as long to perform this task. We decided it
would be fun to help Brian out with his problem, so we came up with an
abstract to solve his
problem. We all agreed that it was a good place to start. After some
time (and a lot of labor) we came up with some code and a paper
based on our work which more closely resembles a collection of notes
than a coherent piece of work. So, we kept working. We have now
arrived at another version of our code. This version works much
better than before; it is better, stronger, and faster. And to this
end, we have written yet another technical paper to describe exactly
what we did and why we did it.
...and what do we learn from this?
Lots! We now know the intimate details of how TCL, DP, and RIVL work
and why performance suffers when certain things are done. Mastering
the black art of TCL programming actually took the smallest amount of
time in comparison to the rest of code. Squeezing the most
performance out of a single-threaded, event-driven environment given a
distributed architecture turned out to be the most time consuming and
difficult part of the project. RPC facilities built into DP allowed
us to easily perform necessary communication between machines to
distribute work requests. The combination of the above scripting
environments helped us to acheive our goals of delivering reliable,
portable, high performance MPEG encoding.
...and how does it tie in with
Multimedia Systems?
Very well! The ability to rapidly compress video using all available
machines makes many new types of video processing possible. It allows
animators and movie editors to view their videos more often, thus
shortening development time. Because our version of parallel MPEG
encoding is an integral part of RIVL, it is extremely easy to use.
...and how do we measure success?
If we are able to compress a large video without becoming old and
gray, we will have succeeded.
...and finally, how are we progressing?
We are finished with the coding state of the project. Currently, we
are working writing a technical report for the project. Hopefully we
can pull the main ideas together better than we did earlier. Some of
things that could be done to make this project better are:
- Make the worker process into a daemon process.
- Add in support for IP Multi-cast.
MULVA - Name of Jerry Seinfeld's girlfriend which rhymes with a
part of the female anatomy.
Courtesy of the "Junior Mint" episode of
Seinfeld.