SLA-Aware Scheduling with Slack-Weighted Prioritization for Multi-Tenant Cloud Orchestration Systems

Abstract

Multi-tenant Software-as-a-Service (SaaS) platforms must satisfy heterogeneous Service Level Agreements (SLAs) while balancing tenant priorities under dynamic load. Existing schedulers such as Earliest Deadline First (EDF) emphasize deadline urgency, while proportional-share mechanisms such as Two-Dimensional Fair Queueing (2DFQ) emphasize allocation fairness; however, neither directly targets SLA compliance and cross-tenant latency disparity within a single interpretable rule.We present SLA-FairSched, a scheduling policy that assigns dynamic priority using a weight-to-slack score, where slack denotes remaining time before a job violates its tenant deadline. The score increases as slack shrinks and remains weight-sensitive when deadlines are not imminent, so SLA compliance and response-time disparity can be balanced in one rule.We evaluate SLA-FairSched using discrete-event simulation with 100 tenants, a 10:1 weight ratio, and 30 runs across system loads up to ρ = 0.95. At high load, SLA-FairSched reduces the tenant-weighted mean SLA violation rate by 18–29% compared to EDF and the HAVEN-style proxy. Although 2DFQ achieves a lower raw violation rate at extreme load, SLA-FairSched improves Jain’s response-time equality from 0.38 to 0.85 at ρ = 0.95, showing lower violation rates need not coincide with equal cross-tenant latency.Slack-weighted prioritization therefore improves SLA outcomes while holding down cross-tenant response-time disparity in multi-tenant cloud orchestration.