Preventing parallel processes from unexpected ineficiencies is a major concern
for constructing multiple user/multiple job environment in distributed memory systems.
Systems can achieve high performance by using shcheduling policies which
reflects resource comsumption states.
For a general environment, which must support concurrent execution of multiple processes,
a way is needed to keep systems' effectiveness when phisical memories are full.
In distributed systems, memory pages can be classified by access frequencies
and required costs for accesses after target pages has been replaced.
Selecting victim pages according to the classification may enhance system performance.
We built a probabilistic model with a concrete memory management scheme
and differntiated memory access costs, and incorporated memory reference frequencies to it.
The paper describes an evaluation of scheduling policies using resource informations for
each process and of page replacement policies under the model.