Synchronizing hypercube networks in the presence of faults

Synchronizing distributed networks allows nodes to share resources efficiently, run synchronous programs and vote on redundant results in fault tolerant systems. Due to the low connectivity of hypercube networks, neither the fault tolerant hardware synchronization schemes, phased locked loops nor multistage synchronizers can be used without adding additional links. We describe a new hardware method developed to synchronize hypercube networks. Our analysis shows that the method can sustain one fault if the connectivity of the hypercube, n, is at least three, and it can tolerate up to m⩾2 Byzantine faults as long as the connectivity of the hypercube, n, is greater than max{2m+1,3m-2). This scheme has been implemented in an ASIC design for a hypercube of dimension five. It will be used in the Proteus parallel computer system to synchronize the circuit switching communication network