Space Sharing Effects on Manycore Systems

Computing hardware has changed from a single core to a multi-core environment, and programming technique have also shifted from a single thread to multi-threaded model. In a multicore environment where resources are limited, time sharing makes multiple application running simultaneously by using resource efficiently. However, in manycore systems, time sharing technique generates kernel noise and it results in a performance degradation. To solve this problem, this paper proposes a space sharing scheme suitable for manycore systems. Space sharing scheme consists of one core-one thread and one core-one memory, ensuring that threads executed without competing kernel resources. Through experiments, we confirmed that space sharing provides scalability compared to time sharing and reduce kernel noise. In addition, we confirmed its performance through a simple HPC application; Edge-detection. We discover that space sharing can create resource waste for I/O intensive applications, which is rather a performance degradation. This problem should be solved through I/O acceleration later.

[1]  Francisco J. Cazorla,et al.  A Quantitative Analysis of OS Noise , 2011, 2011 IEEE International Parallel & Distributed Processing Symposium.

[2]  Adam Wierman,et al.  How to Determine a Good Multi-Programming Level for External Scheduling , 2006, 22nd International Conference on Data Engineering (ICDE'06).

[3]  K. Thompson,et al.  The UNIX time-sharing system , 1978 .

[4]  Robert C. Daley,et al.  An experimental time-sharing system , 1962, AIEE-IRE '62 (Spring).