DYNAMICALLY CONTROLLING FALSE SHARING IN A DISTRIBUTED SHARED MEMORY

Distributed shared memory （DSM） alleviates the need to program message passing explicitly on a distributed memory machine. In order to reduce memory latency, DSM replicates copies of data. Here examined are some current approaches to controlling thrashing caused by false sharing in a DSM. A novel memory consistency protocol （writer owned protocol） is introduced, which detects and eliminates false sharing at run time. In iterative computations where the data is accessed similarly at every iteration, the writer owned protocol can have tremendous benefits because the overhead of eliminating false sharing is incurred only once. Performance results show that the writer owned protocol is often better than the existing approaches.