Search

In this work we examine how the updates addressing Meltdown and Spectre vulnerabilities impact the performance of HPC applications. To study this we use the application kernel module of XDMoD to test the performance before and after the application of the vulnerability patches. The application kernel module is designed for continuous performance monitoring of HPC systems. Here, we tested the performance difference for multiple applications and benchmarks including: NWChem, NAMD, GAMESS, ENZO, HPCC, IOR, MDTest and IMB. The results show that although some specific functions can have execution times decreased by as much as 74%, the majority of individual metrics indicate little to no decrease in performance. The real-world applications show a 2-3% decrease in performance for single node jobs and a 5-11% decrease for two node jobs. For node-counts up to 8 the degradation continues to increase reaching 27% in some cases.

XDMoD, a comprehensive tool for managing computing resources, collects and aggregates numerous metrics that describe computational jobs, resources consumed, wait times, and quality of service. The XDMoD web interface then enables users to chart and visualize these metrics, allowing them to expose and plot related information about their systems via interactive drill-down. An open-source version of XDMoD helps us manage our own center, including several gateways, at University at Buffalo's Center for Computational Research. In this poster, example plots from XDMoD will showcase its current support for gateways, and demonstrate how administrators and end users alike can benefit from its features.XDMoD provides numerous metrics relevant to gateways that describe resource usage and utilization, including number of jobs, CPU hours, wall times, wait times, and job sizes. These metrics are fully integrated with the XDMoD tool, and may be aggregated over any desired time frame. These metrics form the base capabilities of XDMoD. Additional modules further extend XDMoD, providing such functionality as job level performance metrics (Job Viewer). The Job Viewer presents details about a job's executable, its accounting data, job scripts, application information, and timeseries plots of individual metrics such as CPU user, flops, parallel file system usage, and memory usage. These data reflect the details of each job and empower support personnel and resource users to troubleshoot individual job performance or evaluate job efficiency. This poster will display a variety of metrics showcasing the capabilities of XDMoD. It will display aggregated gateway jobs and usage data collected from our own HUBzero gateways, and plotted using our own open-source instance of XDMoD. With this, we hope to introduce the utility of XDMoD for the management of gateways.XDMoD was developed to support NSF's traditional HPC systems, but it offers a boon to the gateways community as well. Using this tool, gateways administrators can explore and visualize data that enables them to manage their existing gateways, plan wisely for future gateway systems, and help end users run jobs more efficiently and make better use of computing resources.

The U.S. National Science Foundation (NSF) has invested heavily in research computing, funding XSEDE to integrate supercomputers with science gateways and datasets for researchers in the U.S. and around the world. It is important to understand how these tools contribute to knowledge, plan wisely for future resource investments, and enable end users to make better use of these resources. Enter XDMoD (XD Metrics on Demand), a comprehensive tool that collects and presents detailed data about resource usage, for program managers, developers, support staff, and end users alike. As XSEDE adds non-traditional resources and diversifies its computational resources, the considerable capabilities offered by XDMoD must keep pace. In this short paper, we introduce XDMoD's current capabilities, describe the state of its support for gateway resources, and outline our plans to further enhance these offerings.