H ARDWARE S OFTWARE D ISTRIBUTED RTOS A RCHITECTURE F OR M ULTIPROCESSOR B ASED R EAL T IME H IGH D ENSITY M EDIA A PPLICATIONS

Abstract

Designing RTOS for multiprocessor based system and partitioning the design into hardware and software is the paradigm for the high-end systems evolving to support computation-intensive applications. In this thesis, RTOS is presented that is amalgam of distributed software scheduler and dynamically scalable hardware switch capable of performing all kernel tasks in real time and at wire speed and consequently, minimizing software related scheduling overheads and optimizing resource utilization. Key components of the research output are software framework for mapping high density media applications on multiprocessor based system, its corresponding multilevel scheduler design and design of programmable hardware switch to optimize resource utilization and enhance system performance. In the developed multiprocessor based system, heterogeneous multiprocessors are arranged as Arrays of inter-connected pipelines giving rise to a modified pool-of-pipelines architecture or two dimensional tightly coupled matrix architecture, having scalable, distributed processing and memory capabilities through a plurality of processing elements. The hardware software distributed architecture improves the application optimization, mapped on this multiprocessor based system, by enhancing execution time and minimizing processor idle time, while reducing design complexity and interface problems. RTOS architecture exploits parallelism both in the system application environment and in the distributed computing occurring in the hardware. Based on the type of service, the processes are divided into sub tasks and distributed onto multiple Processors, which are programmed to operate on a set of data, in a certain sequence and in a particular time slot. - 7 -The sequence is maintained by the instruction format of hardware RTOS and the time slot is predicted by the configurations set by the user. Application pipelining framework is presented in the software part of RTOS to exploit parallelism in media application, based on the concept of components or task model, similar to instruction pipelining in the Processor. To meet the constraints and characteristics of the defined framework, two level scheduling policy is adapted. Fixed- Priority Preemptive scheduling algorithm is used at the First-level. For Second-level various scheduling algorithms have been evaluated and analyzed offline and implemented in the software. Heuristic based algorithm is recommended as Second-level algorithm for media application to maintain QoS. The hardware RTOS, called Switch, is implemented as a programmable processor with dedicated program cache. It performs all RTOS kernel tasks in real time and facilitates inter-processor exchange of data, communication or synchronization, deadlock avoidance and exchange of data with external shared memory, thus, enhancing performance and minimizing Processors idle time. The RTOS supports hardware scalability to cater for high to very high density applications and incorporates scalable scheduling and resource management