Toward inference based distributed processing : analytical modelling and performance evaluation

Abstract EN

Multi-processor distributed systems are very useful for the computation of intensive load.

However, the optimal load fractions allocated to each processor are the main issues that affect the performance of these systems.

Therefore, these load fractions depend on different parameters, such as processing capability for each processor and communication time with each processor.

So, these values must be known before we start to distribute the load.

In this paper, we present a model-based approach to study the performance of multi-processor distributed systems and the different communication times that use an inference message to collect such information.

We notably propose a new model for load distribution over different regions connected through wide area network (WAN).

We mainly focus on the effect of total communication time over the final result.

Performance analyses are evaluated by a simulator, based on C++ programming language that can be an excellent solution to calculate the total finish time, in addition to the limits over the maximum number of regions.