A spatial-temporal parallel approach for real-time MPEG video compression

We present a parallel implementation of an MPEG encoder on the Intel Paragon supercomputer. In our approach, both spatial and temporal parallelism have been exploited, While the Paragon has the computation capacity to encode video sequences faster than real-time, we found that real-time performance cannot be achieved if the input/output techniques are not designed properly. We present several schemes for managing the I/O and show their effect on reducing I/O contention