“Green” system design (whether at the server or datacenter scale) requires a focus on environmental sustainability. Prior studies have focused on operational energy consumption as a proxy for sustainability, but this metric only captures part of the environmental impact–specifically, only the system's impact during its operation. This talk proposes that to understand a system's total impact on environmental sustainability, one needs to examine the entire lifecycle of a system, beyond operational energy to also include material extraction, manufacture, transportation, and recycling. The talk will focus on the two main contributions of this work: First, we proposed a methodology that allows such a lifecycle analysis, specifically providing attribution of sustainability bottlenecks to individual system architecture components such as CPU, DRAM, disk, etc.; second, using this methodology, we compared the sustainability tradeoffs between popular energy-efficiency optimizations–energy proportionality, server consolidation, and low-power embedded hardware (MicroBlades)–across a wide workload space. The talk provides proof that energy-efficiency does not necessarily enable sustainability. Finally, I'll discuss the sustainability bottlenecks and optimizations for future system designs along with some of the current work surrounding system co-designed between architecture, packaging and cooling.
Justin Meza is a graduate student at Carnegie Mellon University in
Pittsburgh, Pennsylvania pursuing his Ph.D. in Electrical and Computer
Engineering, advised by Onur Mutlu. He received his B.S. at
University of California at Los Angeles in Computer Science.
He also works in the Exascale Computing Lab at Hewlett-Packard
Laboratories in Palo Alto, California performing research regarding
energy-efficiency metrics and sustainability practices in computing
platforms from handhelds to datacenters.
Back to the seminar page