Efficient task scheduling for runtime reconfigurable systems

Dynamic Task Distribution Model for On-Chip Reconfigurable High Speed Computing System

“…The MOLEN processor is designed with arbitrary number of processing elements to support both hardware and software tasks. An efficient multitask scheduler [14] proposed for runtime reconfigurable system introduced a new parameter called Time-Improvement as cost function for compiler assisted scheduling models. The Time-Improvement parameter is the combination of reduction-in-task-execution time and distance-to-next-call.…”

“…The Time-Improvement parameter is the combination of reduction-in-task-execution time and distance-to-next-call. The efficient multitask scheduler [14] is demonstrated for the MOLEN polymorphic processor [3] environment where the control of tasks assigned to General Purpose Processor (GPP) and tasks execution is assigned to reconfigurable processing elements. The scheduler in [14] outperforms its contemporary algorithms and accelerates task execution by 4% to 20%.…”

“…The efficient multitask scheduler [14] is demonstrated for the MOLEN polymorphic processor [3] environment where the control of tasks assigned to General Purpose Processor (GPP) and tasks execution is assigned to reconfigurable processing elements. The scheduler in [14] outperforms its contemporary algorithms and accelerates task execution by 4% to 20%. In [15], an online hybrid scheduling model is demonstrated for CPU-FPGA platform where tasks are represented in three categories such as software tasks (ST) executed only on CPU, hardware tasks (HT) executed only on FPGA, and hybrid tasks (HST) executed on both CPU and FPGA.…”

“…The partitioned tasks in Algorithm 2 are further prioritized based on task distribution policy called Minimum Laxity First (MLF) and distributed them onto available resources of RHSCS. There are RLU Implementation Queue and CPU Implementation Queue, in task distribution model, reserved to store the tasks which could be executed on hardcore PE (RLUs) and / * Input: DAGs of application and number of DAG Output: Level annotated tasks of DAGs * / (1) Read number of DAGs (2) for = 1 to number of DAG do (3) Read number of tasks in DAG (4) for = 1 to number of tasks in DAG do (5) Read the number of tasks which depend on task (6) end for (7) Compute Level of the tasks in DAG (8) for = 1 to number of tasks in DAG do (9) while (Level of task > 0) do (10) Assign Level to the task (11) end while (12) end for (13) Sort the tasks in DAG according to the Level assigned (14) Task Resource Mapping (Level annotated tasks of DAG , number of tasks in the DAG ) / * Function of Task Resource Mapping is described in Algorithm 2 * / (15) end for for = 1 to number of Level annotated tasks in DAG do (5) if area of task < size of available RLU then (6) assign to HT Queue (7) else (8) assign to ST Queue (9) end if (10) end for (11) end while (12) Task Distribution (partitioned tasks of DAG, number of tasks in the DAG) / * Function Task Distribution is described in Algorithm 3 * / Algorithm 2: Pseudo code for Task Resource Mapping function.…”

Untitled

“…The MOLEN processor is also incorporated with arbitrary number of programmable units to support both hardware and software tasks. An efficient multi task scheduler for runtime reconfigurable systems [9] proposed a new parameter called Time-Improvement as cost function for compiler assisted scheduling algorithm. The Time-Improvement heuristic is defined based on reduction-in-taskexecution time and distance-to-next-call.…”