Randomized Priority Queues for Fast Parallel Access

Abstract

Applications like parallel search or discrete event simulation often assign priority or importance to pieces of work. An e ective w a y to exploit this for parallelization is to use a priority queue data structure for scheduling the work; but a bottleneck free implementation of parallel priority queue access by many processors is required to make this approach scalable. We present simple and portable randomized algorithms for parallel priority queues on distributed memory machines with fully distributed storage. Accessing O n out of m elements on an n-processor network with diameter d requires amortized time O , d + log m n with high probability for many network types. On logarithmic diameter networks, the algorithms are as fast as the best previously known EREW-PRAM methods. Implementations demonstrate that the approach i s already useful for medium scale parallelism.

Invasive Computing: An Overview

“…For example, diffusion-based load balancing methods [4,1,12] are a simple and robust distributed approach for this purpose. Even centralised algorithms based on global prioritisation can be made scalable using distributed priority queues [13]. Very good load balancing can be achieved by a combination of randomisation and redundancy, using fully distributed and fast algorithms (for example, [14]).…”

Parallel Load Balancing for Problems with Good Bisectors

“…Altough we will show in Section 3.4 that this can be implemented without increasing the asymptotic running-time, it must be expected that substantial communication overhead will occur if many processors want to acquire the identifier of a free processor simultaneously. Further, the second phase of Algorithm PHF requires global communication in each iteration; effectively, the described implementation simulates a specialized parallel priority queue (see [2,9,23] for further information on parallel priority queues) that allows selection of the min[h, f] heaviest remaining subproblems. While this overhead may be small on parallel machines with high-bandwidth and low-latency interconnection networks, it is likely to limit the speed-up achievable with this algorithm in practice on less powerful platforms, such as networks of workstations.…”

A parallel priority queueing system with finite buffers

“…A fresh look at priority queues has appeared in the literature recently. See, for instance, [2,3,21,25]. From what is available in the literature, it is almost clear that standard available methods to solve multi-server queues will not result in a closed form solution for the problem described above.…”

“…Priority queues have been studies extensively and started in the early days of queueing theory. See [4,14,25], for example. A fresh look at priority queues has appeared in the literature recently.…”