Hybrid algorithms for list ranking and graph connected components

Abstract

The advent of multicore and many-core architectures saw them being deployed to speed-up computations across several disciplines and application areas. Prominent examples include semi-numerical algorithms such as sorting, graph algorithms, image processing, scientific computations, and the like. In particular, using GPUs for general purpose computations has attracted a lot of attention given that GPUs can deliver more than one TFLOP of computing power at very low prices.In this work, we use a new model of multicore computing called hybrid multicore computing where the computation is performed simultaneously a control device, such as a CPU, and an accelerator such as a GPU. To this end, we use two case studies to explore the algorithmic and analytical issues in hybrid multicore computing. Our case studies involve two different ways of designing hybrid multicore algorithms. The main contribution of this paper is to address the issues related to the design of hybrid solutions.We show our hybrid algorithm for list ranking is faster by 50% compared to the best known implementation [Z. Wei, J. JaJa; IPDPS 2010]. Similarly, our hybrid algorithm for graph connected components is faster by 25% compared to the best known GPU implementation [26].

Work efficient parallel algorithms for large graph exploration on emerging heterogeneous architectures

“…• On the connected components problem, we get an average 20% improvement over the best known result on an identical platform [6]. A small change to the algorithm can also build a spanning tree of a graph with very little extra time.…”

“…On GPUs, a variant of Shiloach and Vishkin [41] is used by Soman et al [42]. A heterogeneous execution of this algorithm on a CPU+GPU platform with an improvement of 35% on average is shown in [6].…”

Extending the Scheduling Toolkit SEParAT to Support Hybrid Parallel Platforms

“…A single cluster node is defined by the number of CPU cores (line 6), the communication performance of the node-internal interconnection network (lines 7-8) and the compute performance of the processor cores (line 9). Additionally, each cluster node contains a list of zero or more submachines (lines [10][11][12][13][14][15][16][17] where each submachine has a unique identifier and a submachine type, which is used to determine whether a given parallel task is executable on this submachine. The compute performance of a submachine is specified by the average execution time of an arithmetic instruction (lines 12 and 16).…”

“…8 shows an example schedule for a hybrid platform. A schedule consists of communication operations (lines 2-12) and the execution of parallel tasks (lines [13][14][15][16][17][18][19] where each operation has an associated start time and an associated finish time. A communication operation additionally contains the identifiers of the source and target nodes in the underlying task graph and the definition of the source and target execution units.…”

“…In the data parallel case, the input data is partitioned over the available execution units and processed according to the owner-computes rule. Examples for such implementations are the Jacobi method [11], the FFT [12], and the determination of connected components in graphs [13]. The partitioning can be based on profiling information [14], [15] or on static source code analysis in a compiler [16].…”