2026-04-15
Jianbin Qin, Yibin Lin, Wendi Hua, Rui Mao, Chuan Xiao
Published at ICDE 2026
Abstract:
In this paper, we propose Value per Second (VPS), a novel performance metric for OLTP databases that incorporates transactional value into the performance evaluation framework, addressing the weaknesses of traditional throughput- and latency-based metrics. VPS provides a more accurate reflection of real-world business performance. We extend the TPC-C benchmark as an example to illustrate the evaluation insights enabled by VPS and highlight the challenges in achieving high VPS in database systems. In summary, we envision that the next generation of database evaluation should integrate transaction value with system throughput. VPS indisputably better aligns with business objectives and operational significance, offering a more meaningful performance metric for end users. Currently, VPS is in the early stage of development. Several open challenges remain, including cross-domain value modeling, value prediction, and the design of value-aware transaction scheduling algorithms optimized for VPS. We encourage collaboration among researchers across communities to advance this emerging direction.
