<span class="var-sub_title">Power API and Redfish: Standardizing Power Measurement and Control for HPC</span> SC18 Proceedings

The International Conference for High Performance Computing, Networking, Storage, and Analysis

Power API and Redfish: Standardizing Power Measurement and Control for HPC


Authors: Ryan Grant (Sandia National Laboratories), Torsten Wilde (Hewlett Packard Enterprise), James Laros (Sandia National Laboratories), Jeff Autor (Hewlett Packard Enterprise), Todd Rosedahl (IBM), Steve Martin (Cray Inc), Barry Rountree (Lawrence Livermore National Laboratory)

Abstract: The HPC community faces considerable constraints on power and energy of HPC installations. A standardized, vendor-neutral API for power measurement and control is needed for portable solutions to these issues at the various layers of the software stack. In this BoF, we discuss the Power API and Redfish; APIs for measurement and control of power/energy on large systems. The BoF will introduce newcomers to these efforts, differentiate the goals of the two APIs and discuss inter-operability. An interactive panel discussion with experts from involved organizations will facilitate discussions between both API communities with ample time for audience questions and comments.

Long Description: Among the universal challenges that the HPC community faces going forward are constraints on power and energy of HPC installations. However, differences in vendor proprietary interfaces to power measurement and control mechanisms result in duplication of effort and fragment the user base, rendering solutions non-portable between different systems. Moreover, different vendors may be responsible for different components of the same system, including hardware, OS, application-level run time systems, schedulers, and RAS systems. The HPC community has been developing a standardized, vendor-neutral API, the HPC Power API in response to this need. At the same time, the broader commodity market has been developing Redfish, a replacement for IPMI that has some overlap with the Power API. As such, there is great interest in continuing interaction between these two communities towards their common goal of standardized power/energy APIs.

The Power API for HPC is an effort to standardize power measurement and control to address the challenges of energy efficiency and power constraints through a portable API. Initiated by the US DOE and developed by Sandia National Laboratories and interested vendors, the API is now a community effort with participants from industry, national laboratories, government, and academia.

Several previous BOFs at SC14, SC15 and SC17 provided a public forum for discussion of progress on implementations and integration of Power API and Redfish.

While other SC BOFs have be held on power-aware and energy efficient computing, this BOF is the only one focused completely on standardized power/energy monitoring and control APIs. Attendance in previous years has been strong with ~70 people at SC17 and ~55 at SC15 (which was extremely space limited). Since the APIs range from hardware and operating systems to resource managers and administrators to applications, the discussion draws a diverse cross-section of SC attendees. Active work on Power API and Redfish is geographically diverse: in North America, Intel, HPE, Cray, Adaptive, AMD, Qualcomm, IBM, and national labs. In Europe, Bull-Atos and university partners. In Asia, Japan has adopted the Power API via Fujitsu. The range of Redfish adopters further expands this reach, with vendors from all over the world participating. Due to its tradition, scope, and scale, SC is the best venue to bring together the world-wide partners in the Standardizing power/energy monitoring and control APIs and hopefully attract new partners to the effort. The Power API and Redfish are already interoperable (Power API is HPC-centric while Redfish is commodity-centric). Further interaction between the communities will enable greater interoperability and collaboration in the future.

The outcome of the BOF will be a written report summarizing the results of a survey of the attendees, including their responses to the following questions:

-- What API solutions have you deployed on current systems? What do you plan to deploy in the future? -- What areas of interoperability between Power API and Redfish are the most important to you? Are there gaps that need to be addressed?

Attendees will be encouraged to complete a paper survey at the event, with an electronic version also available.


URL: https://eehpcwg.llnl.gov/conf_sc17a-copy.html



Back to Birds of a Feather Archive Listing