Global Shared Memory
HyperShare-based cluster networks operate as global shared memory mapping architectures, meaning that each node in the cluster, along with their internal functional resources such as memory, storage, co-processing and other computing sub-systems, are memory mapped across the cluster. Such global memory mapping scheme gives each cluster node the ability to "see" and directly access memory-mapped resources within any other node across the cluster as if they were its own local ones. This is accomplished by direct, hardware-level Remote DMA (RDMA) access, without having to go through the high latency Remote DMA (RDMA) process of "remote node-to-node data request, wait for remote node acknowledgement and node-to-node RDMA of data with the additional software overhead required for completing the transaction", which is the best that present days cluster network technology has to offer. Simply put, with HyperShare the cluster node-to-node behavioral model changes from "my hardware resources are mine, you have to ask for the permission to access the data and, if I decide so, I will give you permission to access them" to "my hardware resources are yours to access directly and at any time". The net result is that such innovative HyperShare capability enables true cross cluster hardware-level resource virtualization and the major economic and performance benefits that it brings - see the "Hardware Cross -Cluster Level Resource Virtualization" page for details.