As cluster complexity increases, knowing what’s really going on with your application can be a struggle. Recently at the KTH Royal Institute of Technology in Sweden, they decided to see if Allinea Performance Reports could help them with this challenge. So far so good, KTH reports that they are already enjoying increased workload throughput as a result.
Allinea Performance Reports provides a simple report that tells the user how well optimized an application is for the system it is running on, what could be improved, and how it could benefit from running at scale. I/O or networking bottlenecks are instantly identified, and suggestions given on software and configuration changes that will improve performance.
It’s exactly what we wanted – simple and non-intrusive, and quickly identifying potentially problematic applications,” says Erwin Laure, director of the PDC Center for High-Performance Computing. Laure runs the PDC Center for High-Performance Computing in Stockholm, Sweden, providing leading HPC services to Swedish academia as part of the Swedish National Infrastructure for Computing (SNIC), and internationally via the PRACE (Partnership for Advanced Computing in Europe) infrastructure.
The tool is now running on three clusters at the center – one Cray cluster for larger jobs and two smaller Linux clusters. Usage had been reaching 95% over 36,000 cores and it was vital to identify poor performance and resolve issues.
Before using Allinea Performance Reports, we basically looked at applications in a non-systematic way, trying to spot the ones that use most resources, the ones with strange behaviors,” says Laure. “Now, we just run Allinea Performance Reports, identify the problems and then allocate the time and resources that we need.”
In this video, Mark O’Connor from Allinea demonstrates how the company’s Performance Reports software generates easy-to-understand pages that characterize the performance of HPC application runs.