Abstract:

For large-scale system area networks, how to effectively monitor the heavy traffic, detect the performance bottlenecks and locate the potential failure points in order to support the optimization of the network performance becomes a new research topic. We firstly put forward a performance management architecture for the system area network called SNPMA, which adopts the loosely coupled hierarchical structure and realizes the automation and operability of the performance management through the collaboration among different layers. Based on the SNPMA, we also propose a network performance evaluation model called NPEM to resolve the problems like the incapability of properly evaluating the performance conditions for network devices and ineffectively predicting the operational status for the whole network in terms of the large-scale system area network. And then we propose a network performance monitoring method called STM based on the adaptive concurrency strategy, which can dynamically adjust the strategies of collecting performance parameters. In the network environment of Tianhe-2, through monitoring the whole network using the STM, the performance indicators of network devices are effectively evaluated, which verifies the validity of the network performance evaluation model NPEM.

Key words: performance evaluation, system domain, large-scale network, concurrency