Resource Oblivious Sorting on Multicores

Abstract

We present a new deterministic sorting algorithm that interleaves the partitioning of a sample sort with merging. Sequentially, it sorts n elements in O(n logn) time cache-obliviously with an optimal number of cache misses. The parallel complexity (or critical path length) of the algorithm is O(logn loglogn), which improves on previous bounds for deterministic sample sort. Given a multicore computing environment with a global shared memory and p cores, each having a cache of size M organized in blocks of size B, our algorithm can be scheduled effectively on these p cores in a cache-oblivious manner.

We improve on the above cache-oblivious processor-aware parallel implementation by using the Priority Work Stealing Scheduler (PWS) that we presented recently in a companion paper [12]. The PWS scheduler is both processor- and cache-oblivious (i.e., resource oblivious), and it tolerates asynchrony among the cores. Using PWS, we obtain a resource oblivious scheduling of our sorting algorithm that matches the performance of the processor-aware version. Our analysis includes the delay incurred by false-sharing. We also establish good bounds for our algorithm with the randomized work stealing scheduler.