Leveraging Approximation to Improve Datacenter Resource Efficiency

Abstract

Cloud multi-tenancy is typically constrained to a single interactive service colocated with one or more batch, low-priority services, whose performance can be sacrificed. Approximate computing applications offer the opportunity to enable tighter colocation among multiple applications whose performance is important. We present Pliant, a lightweight cloud runtime that leverages the ability of approximate computing applications to tolerate some loss in output quality to boost the utilization of shared servers. During periods of high contention, Pliant employs incremental and interference-aware approximation to reduce interference in shared resources. We evaluate Pliant across different approximate applications, and show that it preserves QoS for all coscheduled workloads, while incurring at most a 5% loss in output quality.

Publication
In IEEE Computer Architecture Letters (CAL), 2018