Yang and Chien look at the opportunities and costs of using stranded power (SP) at green generating centers for supercomputing and allied computer resources. They “define [SP] as offered power generation with no economic value, thus including both curtailment and power delivered with zero or negative [locational marginal price, LMP].” The paper is an extension of the authors’ work on exclusively powering supercomputers with SP, called zero-carbon cloud (ZCCloud) [1].
To begin with, the authors discuss the nature of the power grid and energy markets, followed by an SP quantification method, data center costs, and modeling. They then review other works in this area and discuss the ZCCloud methodology. Next is a discussion on the productivity of intermittent resources, which centers on a simulation of high-performance computing (HPC) workloads using a daily periodic model, and uptime intervals for a wind generation site. The periodic model is based on the Argonne Leadership Computing Facility (ALCF) Mira system for traditional data centers, with 49152 nodes, 786432 cores and 768 terabytes (TB) of memory. ZCCloud is connected to the data center (Ctr+Z); ZCCloud extensions are also covered. The intermittent resources are modeled first by a fixed daily cycle, and subsequently by a duty factor. The authors then discuss the total cost of ownership (TCO) of traditional and ZCCloud systems, find thating (1) productivity is a function of the reliability and availability of resources, and (2) optimizing data center locations cannot reduce SP.
The paper closes with a commentary on future work. It has 92 references and should interest researchers in the energy sector.