Conclusions

 

The following conclusions summarise the findings of this chapter. They relate directly the the hypotheses from chapter 3 and the results are only valid for the distribution of heavyweight processes between a small number of hosts. Chapter 10 applies these specific results to more general issues in load distribution.

• The process mix hypothesis is supported for both initial placement and process migration:

  The reduction in average response times show that the process mix hypothesis is supported for initial placement, and strongly supported for process migration over a wide variety of workloads.

• The prediction accuracy hypothesis is supported:

  Good performance was gained using a weak predictor, supporting the hypothesis.

• The initial placement hypothesis is supported:

  Initial placement is as good, if not better than migration if the sole aim is to improve the performance of a distributed system. The extra effort in implementing a full process migration system is not reflected in the resulting performance gain over initial placement for the process based system model studied.

  Thus process migration of heavyweight processes appears to be more suitable for other requirements, such as migrating servers or user shells for reliability and avoidance of hot points, or migrating to enforce ownership rights, and other such policy oriented decisions.

• The combination hypothesis may apply during heavy workloads: However, in general it offers no real advantage over initial placement or process migration.