Fine-grained Application Energy Modeling

Project Information

Discipline

Computer Science (401) 

Orientation

Research 

As with performance, energy-efficiency is not an attribute of a compute resource alone; it is a function of a resource-workload combination. The operation mix and locality characteristics of the applications in the workload affect the energy consumption of the resource. Data locality is the primary source of variation in energy requirements. The major contributions of this work include a method for performing fine-grained DC power measurements on HPC resources, a benchmark infrastructure that exercises specific portions of the node in order to characterize operation energy costs, and a method of combining application information with independent energy measurements in order to estimate the energy requirements for specific application-resource pairings.

This work represents a step toward being able to better evaluate the energy-efficiency of HPC compute resource prior to acquisition. This is the fist time, to the the authors knowledge, that the energy-efficiency of compute resources can be modeled for specific applications.

Any entity that is responsible for HPC acquisitions has to take energy budgeting into account. Developing methods and tools for estimating energy costs and design trade-offs has potential uses across several government agencies.

Project Contact

Project Lead

Catherine Olschanowsky (cmills) 

Project Manager

Catherine Olschanowsky (cmills) 

Resource Requirements

Hardware System

• sierra (IBM iDataPlex at SDSC)

 

Similar to any purchasing decision, high performance computing (HPC) resource acquisition decisions involve balancing workload requirements against cost. These requirements focus primarily on computing performance, ideally expressed in terms of application performance. In the past, HPC facility cost was dominated by hardware. Less expensive commodity components being incorporated into larger and larger systems combined with the rising costs of power has elevated energy costs to an equal level. As a result HPC design and acquisition decisions are becoming more complex; the value of a resource is now a function of performance, hardware cost, and energy efficiency.Our research has found a direct correlation between the memory behavior of an application and the energy-efficiency of that application on a given resource. Performance modeling combined with resource energy profiles is able to utilize this correlation to predict the performance and energy-efficiency of an application on a given resource. This ability provides valuable information to the acquisition process as well as design and configuration decisions.The energy-efficiency predictions require that an energy profile of the target resource be obtained. The profiles have been designed and tested on a very limited number of resources. The initial results are very promising; predictions have been within 8% of measured energy consumption. The challenge ahead involves insuring that the energy profile methodology can be expanded to a general set of hardware and further improving and verifying the process.The PMaC team is interested in using the proposed Grid Testbed to further develop the energy profiles and benchmarking techniques required for energy-efficiency modeling. In order to generalize the modeling approach we require exclusive access to a variety of hardware configurations. Resource energy profiles characterize the relationship between data movement and energy consumption. In order to collect that information we instrument the hardware with a power monitoring harness. Specially designed benchmarks are run on the system while taking power measurements creating the energy profiles.

Between 2 and 4 nodes of Sierra will need to be reserved during measurements. The measurements take typically between 1 and 3 weeks.

Project Timeline

Submitted

11/23/2010 - 16:05 

Completed

12/31/2010