Bashroush et al. 2022, refurbished vs new servers

Disposition: stated. Read 2026-10-14 in the browser (full text on the IEEE Computer Society page): https://www.computer.org/csdl/journal/su/2022/01/09246737/1oqGe9mvgcg

What: Rabih Bashroush (Uptime Institute research director, UEL) with a UEL/Techbuyer partnership. SERT benchmarks of an HPE DL380 Gen9 in 22 configurations, plus refresh-payback modelling on SPECpower data for dual-socket servers 2010-2019. Findings: an all-refurbished server scored the same as an all-new one of the same model (SERT 13.7 vs 13.8, not significant); reliability of refurbished servers is close to new because failure-prone parts (disks, PSUs) get replaced. Gains per generation have shrunk: moving to the latest interval cut energy only ~16%, and idle power has risen, which hurts at typical ~25% utilisation. Payback: refreshing servers older than ~6 years with refurbished ones (~$1,200, against ~$2,800 new) can pay back in about a year; refreshing servers younger than 5 years pays back poorly whether new or refurbished. Caveat: refurbished results are sensitive to the quality of the refurbishment. Companion figure: Uptime/EURECA (300 European data centres): servers older than five years were 40% of the fleet, 66% of IT energy and 7% of compute (Uptime Institute 2020, search text: https://uptimeinstitute.com/uptime_assets/80ae92ca9b8dfa363a077cb537f51870777499a39218906efc6d4e37e28ac3a0-beyond-pue-tackling-its-wasted-terawatts.pdf). Used on: hosting-cost-claim, trail-2026-10-14.