Graph Problems Performance Comparison Using Intel Xeon and Intel Xeon-Phi
While most modern well known performance benchmarks for high performance computers focused mainly on the speed of arithmetical operations, the increasing amount of nowadays problems depend also on the speed of memory access. This aspect is becoming crucia
- PDF / 144,248 Bytes
- 11 Pages / 439.37 x 666.142 pts Page_size
- 24 Downloads / 173 Views
ct While most modern well known performance benchmarks for high performance computers focused mainly on the speed of arithmetical operations, the increasing amount of nowadays problems depend also on the speed of memory access. This aspect is becoming crucial for all data driven computations. In this paper, two benchmarks focusing on the speed of memory access are examined. The first examined benchmark is well known Graph 500. This benchmark was developed in order to measure the computers performance in memory retrieval using the Breadth First Search algorithm on randomly generated graph. The second benchmark uses the real world data set (Czech Republic traffic network) as an input graph on which the betweenness centrality algorithm is performed. Both of these benchmarks were tested on SALOMON cluster comparing performance on both Xeon processors and Xeon-Phi co-processors. Obtained performance results were analyzed and discussed at the end of the paper. Keywords Graph 500 ⋅ Betweenness centrality ⋅ Benchmark ⋅ Parallel computing ⋅ Xeon ⋅ Xeon-Phi
J. Hanzelka (✉) ⋅ R. Skopal ⋅ K. Slaninová ⋅ J. Martinovič ⋅ J. Dvorský IT4Innovations, VŠB – Technical University of Ostrava, 17. listopadu 15, 708 33 Ostrava, Poruba, Czech Republic e-mail: [email protected] R. Skopal e-mail: [email protected] K. Slaninová e-mail: [email protected] J. Martinovič e-mail: [email protected] J. Dvorský e-mail: [email protected] © Springer Nature Singapore Pte Ltd. 2017 R. Chaki et al. (eds.), Advanced Computing and Systems for Security, Advances in Intelligent Systems and Computing 567, DOI 10.1007/978-981-10-3409-1_5
73
74
J. Hanzelka et al.
1 Introduction The significant amount of nowadays computations is becoming data driven. Therefore, a computer ability to fast memory access appears more and more crucial because of the requirements of data processing and analysis [2]. The important area of such computations are big graph problems which are mostly used in order to model large real life networks, i.e. traffic networks, social networks but also in areas like the modeling of connections in the human brain [6]. For example, one of the used algorithms in area of traffic networks is betweenness centrality described in this paper. The general assumption of betweenness centrality in traffic routing is that the most frequently used roads can be considered as bottlenecks [5]. The main goal of betweenness centrality is to identify such bottlenecks in a traffic network represented by oriented weighted graph. This way, betweenness centrality can help to optimize routing since it allows us to monitor how bottlenecks change during the day. Since betweenness centrality is based on the shortest path search, it can also be used to monitor and examine bottlenecks behavior based on wide number of factors. Besides the distance, graph weights could reflect actual weather conditions, speed, time of the day, etc. Due to these reasons, it becomes highly important to measure and examine the performance of graph algorithms. The authors are the members
Data Loading...
