SDFMapOpt
SDF graph mapper using dynamic scheduling policy
About
- In order to execute a SDF graph on multi-core systems, two problems should be resolved; mapping and scheduling. SDFMapOpt assumes static mapping and dynamic scheduling policy and determine node to processor mapping which minimize resource overhead considering processor cost and total buffer size requirement under given throughput constraint.
- SDFMapOpt uses dynamic scheduling policy which has several benefits over static scheduling approach which is most previous researches assumed. By using dynamic scheduling, the effect of unfolding can be achieved naturally and this can improve the system performance. In addition, since we just need to set priority onto nodes to schedule, it is easier to represent the schedule and uses less memory than static schedule. And if there are variations of execution time or communication delay at run time, the dynamic schedule can response more efficiently than static shcedule.
- Since the mapping and scheduling problem is NP-hard, SDFMapOpt uses MOEA to optimize processor cost and total buffer size simultaneously.