| 摘要 |
Methods for optimizing memory unit usage to maximize packet throughput for a multiprocessor multithreaded architecture. One method employs a first phase of a software algorithm to allocate data structures to memory units in which the data units are stored and accessed during processing operations. The allocation is such that the data structures are allocated to memory units having lower latencies while satisfying capacity and bandwidth constraints for the memory units. A second phase of the algorithm may be employed to tune the allocation, wherein the performance level of an initial allocation and subsequent reallocations are simulated for an environment in which the memory units and data structures are to be implemented. From the simulation, the allocation providing the best performance level is selected. The simulated environment may include network processor unit (NPU) environments, with the performance level comprising a measure of packet throughput.
|
