Scheduling Streaming Applications on a Composable Multi-Processor System

Scheduling Streaming Applications on a Composable Multi-Processor System

Leegwater, A.E.C.

Cotofana, S. (mentor)

Electrical Engineering, Mathematics and Computer Science

Microelectronics & Computer Engineering

Master Embedded Systems

2010-10-28

For real-time streaming applications such as video decoding, the rate of the application is very important. To fully use the available resources of an multiprocessor platform, thus achieve a high rate, the applications have to be scheduled efficiently. We assume a composable multiprocessor, which means that applications on the platform are independent both in functional and temporal behaviour. Prior work provides composability for the processor utilising two levels of scheduling: composable application-level scheduling with Time Division Multiplexing (TDM) and task-level scheduling with an arbitrary scheduler. In this thesis we are concerned with efficient task-level scheduling. We compare two existing schedulers: Time Division Multiplexing (TDM) and Credit-Controlled Static-Priority (CCSP). TDM is simple and couples latency and rate; CCSP originates from memory controllers and decouples latency and rate. To compare these schedulers we create a dataflow graph of the scheduled application and calculate the parameters that result in the maximum rate. Then these parameters are assessed in experiments on a MPSoC implemented on an FPGA. From the results of our investigation we can draw two important conclusions. The first is that the formal models of the schedulers lack accuracy in predicting the latency of a task. This is visible in our experiments with buffer bounded applications. The second conclusion is that CCSP schedules tasks more efficiently than TDM. When the period for TDM is exactly proportional to the execution times of the tasks and the application is not buffer bound, then TDM can equal the performance of CCSP.

composable
scheduling
multi-processor
TDM
CCSP

http://resolver.tudelft.nl/uuid:6c779496-a885-48e4-afef-d97704595159

2010-11-03

Student theses

master thesis

(c) 2010 Leegwater, A.E.C.