BALM: QoS-Aware Memory Bandwidth Partitioning for Multi-Socket Cloud Nodes

The recent emergence of novel hardware-based resource partitioning mechanisms has unveiled the opportunity for a new generation of QoS-aware resource allocation approaches for workload consolidation. Still, to the best of our knowledge, existing proposals are, by design, not tailored to the growing prevalence of multi-socket systems in contemporary warehouse-scale data centers. We propose BALM, a QoS-aware memory bandwidth allocation technique for multi-socket architectures that combines commodity bandwidth allocation mechanisms with a novel adaptive cross-socket page migration scheme. Our experimental evaluation with real applications on a dual-socket machine shows that BALM can overcome the efficiency limitations of state-of-the-art. BALM can ensure marginal SLO violation windows while delivering up to 87% throughput gains to bandwidth-intensive best-effort applications when compared to state-of-the-art alternatives.