This paper investigates the problem of network partitioning in distributed shared memory (DSM) systems. Network partitioning in a DSM system will severely reduce the availability of shared pages. We propose a partition-processing protocol for DSM systems, called one-copy read and nocopy write protocol, to allow that shared pages can be optimistically accessed in each partition without taking care of the consistency with other partitions. However, the proposed protocol may introduce the problem: inconsistency between partitions. To resolve the inconsistency problem, we employ a coordinated checkpointing technique to reconstruct the consistency between partitions. Finally, we perform mathematical analysis to evaluate the availability of the proposed protocol.