For the most recent entries see the Petri Nets Newsletter.

Performance balancing in multithreaded multiprocessor systems.

Zuberek, W.M.; Govindarajan, R.

In: Proc. 4th Australasian Conf. on Parallel and Real-Time Systems (PART'97), 29-30 September 1997, Newcastle, Australia, pages 15-26. 1997. Available at ftp://ftp.cs.mun.ca/pub/publications/97-PART.ps.Z.

Abstract: Fine-grained multithreading is an architectural approach for tolerating long, unpredictable communication latency and synchronization delays in multiprocessors systems. Instead of waiting for the result of a long-latency operation, the processor performs context switching, i.e., it suspends the current thread and switches to another ready thread if such a thread is available. The performance of a multithreaded multiprocessors system depends upon a number of parameters, which, if not selected properly, result in system bottlenecks and limit the performance of the system. This paper derives simple conditions under which a multithreaded multiprocessor system is balanced, so it does not contain bottlenecks. The balancing conditions indicate when multithreading is capable of tolerating long latencies. This, in turn, provides useful guidelines for appropriately choosing the parameters of the multithreaded architecture so as to achieve high performance. Some effects of non-balancing on the performance of the system are also discussed.

Keywords: multiprocessor architectures, multithreaded architectures, performance balancing, timed Petri nets.


Do you need a refined search? Try our search engine which allows complex field-based queries.

Back to the Petri Nets Bibliography