A Case Study in Pipeline Processor Farming: Parallelising the H.263 Encoder
This paper describes the parallelisation of the H.263 hybrid video encoder algorithm based upon a pipelines of processor farms (PPF) paradigm. In addition, a data-farming template, which can be very useful for several image coding algorithms, was incorporated in the PPF model. A variety of parallel topologies were implemented in order to obtain the best time performance for an eight processor distributed-memory machine. Results show that, due to communication overheads and algorithm constraints, the speed-up performance is below the value predicted by static analysis. However, the design examples indicated how to modify the PPF methodology in identifying those algorithm components which restrict scaling performance. The paper highlights the problems associated with the parallelisation of sequential algorithms and emphasises the need for generic tools to facilitate such conversion.
KeywordsExecution Time Pipeline Stage Passing Pointer Parallel Topology High Communication Bandwidth
Unable to display preview. Download preview PDF.
- “ITU-T Recommendation H.263”, Technical Report, 1995.Google Scholar
- D. Bailey, M. Cressa, J. Fandrianto, D. Neubauer, H. Rainnie, Ch-Sh. Wang, “Programmable Vision Processor/Controller for Flexible Implementation of Current and Future Image Compression Standards”, IEEE Micro, pp. 33–39. Oct. 1992.Google Scholar
- Transtech Parallel Systems Corporation “The Paramid Users Guide”, 1993.Google Scholar
- M. Atkins, “Performance of the i860 Microprocessor”, IEEE Micro, pp. 24–78, Oct. 1991.Google Scholar
- M. Fleury, H.P. Sava, A.C. Downton and A.F. Clark “Designing and Instrumenting a Software Template for Embedded Parallel Systems”, In this volume.Google Scholar