Optimizing virtual machine scheduling in NUMA multicore systems

Abstract

An increasing number of new multicore systems use the Non-Uniform Memory Access architecture due to its scalable memory performance. However, the complex interplay among data locality, contention on shared on-chip memory resources, and cross-node data sharing overhead, makes the delivery of an optimal and predictable program performance difficult. Virtualization further complicates the scheduling problem. Due to abstract and inaccurate mappings from virtual hardware to machine hardware, program and system-level optimizations are often not effective within virtual machines.We find that the penalty to access the "uncore" memory subsystem is an effective metric to predict program performance in NUMA multicore systems. Based on this metric, we add NUMA awareness to the virtual machine scheduling. We propose a Bias Random vCPU Migration (BRM) algorithm that dynamically migrates vCPUs to minimize the system-wide uncore penalty. We have implemented the scheme in the Xen virtual machine monitor. Experiment results on a two-way Intel NUMA multicore system with various workloads show that BRM is able to improve application performance by up to 31.7% compared with the default Xen credit scheduler. Moreover, BRM achieves predictable performance with, on average, no more than 2% runtime variations.

When eXtended Para - Virtualization (XPV) Meets NUMA

“…Disco hides the NUMA topology, which makes this solution inefficient for an SRL that implements its own NUMA policy. Similarly, Rao et al [43], Wu et al [53], Jaeung al. [27] and Liu et al [35] proposed new heuristics to place or to migrate the memory or the vCPUs in order to efficiently use the NUMA architecture.…”

Performance comparison of multi-container deployment schemes for HPC workloads: an empirical study

“…Consequently, application partitioning comes together with the need to schedule the resulting virtual machines in the NUMA platform so that each of them optimizes its memory access locality [7,31] or the access to local I/O devices [5]. Cheng et al [7] presented a user-level scheduler that periodically adjusts the placement of virtual machines aiming for local node execution, that is, the VCPUs of a virtual machine are running on one NUMA node and its memory is also located on the same NUMA node.…”

“…Cheng et al [7] presented a user-level scheduler that periodically adjusts the placement of virtual machines aiming for local node execution, that is, the VCPUs of a virtual machine are running on one NUMA node and its memory is also located on the same NUMA node. Rao et al [31] proposed a load balancing algorithm to determine the optimal VCPU-to-core assignment by dynamically migrating VCPUs to minimize the penalty to access the uncore memory subsystem.…”

Convergence of high performance computing, big data, and machine learning applications on containerized infrastructures

“…Consequently, application partitioning comes together with the need to schedule the resulting virtual machines in the NUMA platform so that each of them optimizes its memory access locality [148] [23] or the access to local I/O devices [14]. Cheng et al [23] presented a user-level scheduler that periodically adjusts the placement of virtual machines aiming for local node execution, that is, the VCPUs of a virtual machine are running on one NUMA node and its memory is also located on the same NUMA node.…”

“…Cheng et al [23] presented a user-level scheduler that periodically adjusts the placement of virtual machines aiming for local node execution, that is, the VCPUs of a virtual machine are running on one NUMA node and its memory is also located on the same NUMA node. Rao et al [148] proposed a load-balancing algorithm to determine the optimal VCPUto-core assignment by dynamically migrating VCPUs to minimize the penalty to access the uncore memory subsystem.…”