BTQoS: A Tenant Relationship-Aware QoS Framework for Multi-tenant Distributed Storage System

Quality of Service (QoS) is widely applied in multi-tenant distributed systems to meet tenants’ Service Level Objective (SLO). Previous work ignored the relationship between tenants, making it challenging to achieve hierarchical resource management of tenants. This work proposes BTQoS, a tenant relationship-aware QoS framework for multi-tenant distributed storage systems. BTQoS introduces behavior and behavior tree abstractions to manage tenants’ SLOs and relationships. BTQoS uses a two-level scheduling scheme to implement SLO enforcement, resource adaptation, and skew avoidance under dynamic workloads. In addition, BTQoS’s two-level scheduler decouples tenant resource allocation and cluster resource adaptation, allowing the SLOs supported by BTQoS to be expanded with low development costs. We implement and evaluate BTQoS in HDFS. The experimental results show that BTQoS can hierarchically manage and enforce tenants’ SLOs at low overheads and provide skew avoidance.