Parallel processing on parallel/distributed systems is showing greater availability as network-surrounding micro-electronics evolves and many optimization mechanisms are realized. This paper describes a kernel-level scheduling method to build a general purpose parallel OS on NUMA-type parallel machines. In a distributed memory environment, constructing a multi-user/multi-job world without decreasing high efficiency of parallel applicatoins can be achieved by managing resource informations and scheduling according to them. We simulated on a simplified model and evaluated four scheduling methods.