<#3028#>#tex2html_wrap4622#<#3028#>
It is clear from the literature that there is an order:
process placement ;SPM_lt; load sharing ;SPM_lt; load balancing
As this order is traversed from left to right:
- There is an increase in the likely benefits.
- There is an increase in the fairness of the system.
- There is an increase in the complexity of implementation.
- There is an increase in the difficulty of analysing the stability of
the system.
All approaches are better than none, and it is a policy matter to decide
whether the complexity of implementing a more dynamic system is worth the
effort. There is a measure of unresolved controversy in the area within the
literature; contradictory results have been obtained as a result of using
different metrics and techniques for measuring effectiveness. Nevertheless, it
is not unreasonable to suppose that a simple approach, perhaps utilising only a
single load indicator variable, perhaps without information exchange between
nodes, will give most of the benefits to be had.
Programmer controlled migration of executing processes has advantages
which are not dependent on the particular approach to load distribution chosen.
For example, migration may allow data reduction to be performed on telemetry
data.
<#3029#>#tex2html_wrap4624#<#3029#>