Design Of A Dynamic Scheduling Supported Distributed OS For MPSocC

It is difficult to manage the global resource because of complexity for the MPSoC systems. The key requirement of effective utilization of MPSoC and speedup execution of parallel programs relies on management of the resource by OS for the MPSoC systems, so it's necessary to study OS for MPSoC. The asymmetric Multi Processing model is applied to design a distributed OS for a NoC-based MPSoC, in which nodes are divided into Control node and Operation nodes. A costuming Linux runs on the control node as the interface between end-users and the hard-subsystem, while an extended uC/OS on each operation node to get high performance and support execution of parallel applications.There are two modules on the control node. The resource stat module creates multiple processes to collect resource information and stores these in the stat pool which is constituted of shared memory. Dynamic scheduling technique is adopted to assign tasks to operation nodes by the scheduling module. In addition, the stat module automatically establishes the stat environment and provides required mapping file for the two modules after reading the network configuration file so that the system will hold high scalability.The uC/OS is replanted to ARCA3 and task switch, interrupt handle, network interface driver, and dynamic memory management are implemented. In the OS extension layer, the stat task collects required information determined by the configuration send from control node. By setting the mapping result at the initialization stage, the stat task and load task can communicate with the control node accurately regardless of the scale of the network. Then the load task gets the mapping result, uses paging system and parameter passing to create, load and execute MMPI tasks assigned by control node.The control node cooperates with operation nodes to schedule, assign, load and execute parallel programs. The distributed OS is validate, a test case with four tasks executes on a three-core platform, which includes one control node. The result proves the validity of the OS function, and the cost of every stage such as scheduling, assigning, loading and executing is evaluated, then the influence to the cost of each stage of the system size and application size is estimated. The distributed OS can be used in the embedded systems and to evaluate dynamic scheduling algorithms.