| OCR Text |
Show CHAPTER 4 CONCLUSION AND FUTURE RESEARCH An applicative system is one of the most promising candidates toward aggregating a large number of processors to achieve high performance computation. This dissertation has studied two important subjects in this class of large-scale multiprocessor systems. The load balancing problem concerns the upward scalability, and the fault tolerance deals with the graceful downward scalability issue. The load balancing problem of applicative systems is characterized by a large number of processors and spontaneous generation of numerous concurrent tasks. Consequently, any balancing scheme requiring global actions seems impractical when the system scales up. Spontaneous task generation also makes it difficult to prenegotiate a balanced distribution beforehand. An efficient and elegant load balancing scheme, called the gradient model, is devised. The model is based on a demand-driven principle which requires the underutilized processors to dynamically initiate the load balancing requests. A system-wide gradient surface is formed as a result of these requests. The nearest overloaded processor responds to the request by migrating an unevaluated task down the gradient surface to~ard the underutilized processor. The gradient surface serves as both a load request and load migration pointer. A global balance state is achieved by successive approximation of many localized balances. The concept of saturation is coined. When all processors are fully utilized, further load migration becomes futile. Saturation is easily detected in the gradient model and the futile load migration is avoided. A pilot SIMULA s1mulator has been designed to compare the effectiveness |
