TxComm: Transforming Stream Communication for Load Balance, Efficiency, and Fault-tolerance in Networks-on-Chip

Recent work has examined using application-specific knowledge of streaming communication to optimize network routing (for throughput/performance) and/or design (for simpler hardware). However, previous techniques have assumed that the communication streams are directly mapped to networks-on-chip. In contrast, this paper explores the use of communication transformations (TxComm) to achieve (1) higher throughput via better network load balance, (2) more efficient network utilization, and (3) better fault-tolerance, while retaining the communication semantics of the original streaming application. Specifically, we propose two transformations: stream fission and stream fusion. (While fission and fusion transformations have been applied to computation in streaming programs, we are the first to propose fission and fusion transformations for stream communication.) Stream fission splits streams of communication to multiple streams that may be routed over independent network paths to achieve better network load balance. Stream fusion targets multicast communication and fuses multiple streams to effectively capture the well-known benefits of tree-based multicast, which include more efficient link utilization. Both techniques can be integrated in an integer linear program formulation that executes at compile time. Another key component of TxComm is the use of free routing which serves two key purposes. First, it boosts the performance of fission and fusion. Second, it enables application-specific fault-tolerance. Evaluations with a suite of StreamIT benchmarks show that TxComm achieves significant performance improvement over prior application-specific (non-transformed) routing techniques. On the fault-tolerance front, TxComm achieves similar performance as a fault-free base case even when as many as 10% of links are faulty.