Recomputing static assignment is a very interesting area in parallel computers. The abstraction here is that we want to modify our static assignment to reflect the changing problem. The implementation like we have learned in this class many times depends on the problem and the constraints of the system.
For instance where does the line of recomputing assignment cross the line of it being faster to not have to deal with that communication and overhead? Another question to ask is what work can we reuse from people who design distributed systems as this is a problem they have evolved many solutions to. How much of that applies to our system constraints.
Recomputing static assignment is a very interesting area in parallel computers. The abstraction here is that we want to modify our static assignment to reflect the changing problem. The implementation like we have learned in this class many times depends on the problem and the constraints of the system. For instance where does the line of recomputing assignment cross the line of it being faster to not have to deal with that communication and overhead? Another question to ask is what work can we reuse from people who design distributed systems as this is a problem they have evolved many solutions to. How much of that applies to our system constraints.
This paper goes into detail of 5 techniques to achieve reassignment in parallel computers. http://www.ics.uci.edu/~cs230/reading/parallel.pdf