Total Cost of Ownership and HPC System Procurement
Authors: Gert Svensson (KTH Royal Institute of Technology)
Abstract: The predominant goal for procurement of HPC systems is to identify the optimal solution to both technical and financial targets that maximizes the contribution of that system to the organization’s mission. Beyond the acquisition cost of the system, it is also important to consider the total costs of ownership, including the improvements necessary to host the system, the infrastructure that supports its operation, and the significant operational costs associated with that new HPC system. HPC leaders from across the globe will discuss and debate key procurement requirements and lessons learned that can contribute to greater efficiency and reduced operational costs.
Long Description: Over the past 10 years, operational and infrastructure costs of HPC have become important considerations during system procurement and this trend will only continue as we move towards ever more powerful computers.
Perhaps as few as ten years ago, the operational and infrastructure costs for new HPC systems were largely ignored as part of the procurement phase of an acquisition. As a wide variety of factors drive new HPC systems to become larger, denser, and hotter, their impact on facility infrastructures and operational costs increases and suggests integration of these aspects during the procurement of a system. Whilst such considerations are of paramount importance in the procurements for the largest systems that will seek exascale performance targets, the underlying requirements can lead to greater energy efficiency and important reductions in infrastructure and operational costs for the larger community as well.
The goals of this BOF are to:
• promote dialogue between leadership supercomputing sites that have made an explicit effort to include consideration of these costs in their procurements,
• promote dialogue between supercomputer sites and leading vendors on how to use these costs in procurements in an efficient way,
• capture lessons learned and best practices for the audience and also for publication to a broader audience,
• solicit audience feedback and participation in the discussion.
The Partnership for Advanced Computing in Europe PRACE has recently done a study of Total Costs of Ownership (TCO) principals, including using TCO in procurements. The results from a review of the distribution in size of TCO components at major European sites as well as what those sites has worked and plan to work with to lower the costs will also be presented in a white paper (will be published in December 2017). A preview of the results is shown in this BOF. Prace has also published research on Procurement and commissioning of HPC services.
The Energy Efficient HPC Working Group (EE HPC WG) has published a document on Energy Efficiency Considerations for HPC Procurements. Lessons learned and best practices from this BoF would be captured in an updated version of that document. https://eehpcwg.llnl.gov/pages/compsys_pro.htm
This BoF was held in SC16 and was inspired by a panel discussion held at SC15. Both of these sessions were well attended. This BoF is intended to extend and elucidate those discussions. The main extension is including the PRACE effort along with that of the EE HPC WG.
Panelist presentation (3 minutes each):
What best practices can you recommend including in an HPC system procurement that considers infrastructure and operational costs? Do you have any lessons learned?
Panelist pre-set questions:
What are the relative weights between capital and operational expenses in HPC procurements today? How is this trending? Do you see a cross-over between capital and operational expenses anytime within the next decade?
Do you see shifts in technology or environmental factors that will affect infrastructure and operational considerations for procurement within the next 5 years? (e.g., renewables, regulations)
Conference Presentation: pdf
Birds of a Feather Index