Measuring Energy Usage in Parallel Software/Hardware: Adept session at ISC'14
The Adept project is running a Birds-of-a-Feather session (BoF) on 'Measuring Energy Usage in Parallel Software/Hardware' at the upcoming ISC'14 conference. The session will present early results in measuring the power usage of parallel software and hardware, looking at three strands in particular:
1. Adept is developing benchmarks to help attribute power usage to specific components in a parallel application. We will present our advances in getting accurate and meaningful measurements (as well as the challenges faced), and discuss the value of the information that can be gained.
2. To measure the power usage of parallel software to a high resolution, we need to either instrument the hardware or rely on vendor-exposed energy usage information. We will discuss our methodology, and compare accuracy and reliability of different measurement techniques.
3. We will consider 'performance vs. energy usage' metrics, which are important for directly comparing different types of software applications running on different hardware architectures, and to quantify the performance/energy trade-off.
The session is aimed at :
- Software developers and scientists with interest in the energy usage of parallel (HPC) code on a range of hardware platforms
- Hardware designers/vendors, particularly those with an interest in energy efficiency
- Tools/compiler/runtime systems developers with an interest in exploiting power measurement information in their products.
The BoF will take place as part of ISC'14 in Leipzig on Tuesday 26th June, 9:00-10:00.
For more information, see the webpage for this BoF session.
The Adept project
Adept is addressing the challenge of energy-efficient use of parallel technologies. It will build on the expertise of software developers from high-performance computing (HPC) to exploit parallelism for performance, and on the expertise of Embedded systems engineers in managing energy usage.
The project is developing a tool that will guide software developers and help them to model and predict the power consumption and performance of parallel software and hardware. Adept is also investigating the implications of parallelism in programming models and algorithms, as well as choice of hardware, on energy and power consumption.