چكيده به لاتين
Reconfigurable systems have been known for their flexibility and their strength in
hardware execution in different fields of application. They can be utilized in a wide area
of applications due to their adaptability in various conditions. However, managing the
resources besides executing the tasks in the lowest time is one of the main challenges in
these systems. Many factors affect the performance of the system such as execution time,
communication cost, and required hardware resource. In this work, we propose a geneticbased scheduling algorithm considering execution time, communication cost and resource
utilization of each task. In order to evaluate the proposed algorithm, a fitness function is
represented and tested with a selector function to verify the mentioned function. Several
experiments have been done on randomly generated DAGs and also a set of real
applications DAGs to completely evaluate the proposed method. Moreover, experiments
have been done based on two topological features of DAGs, the number of level one
parallel tasks and critical path length. Moreover, we define a working area for the
proposed method based on the topological features. The results were more satisfying in
the DAGs with more level one parallel tasks and shorter critical paths. The results shows
11% and 23% fitness improvement outside and inside of the working area, relatively.
Keywords: Reconfigurable Systems, Task Scheduling, Resource Allocation, Multiobjective optimization, Evolutionary Algorithms
