Cloud-scale VM-deflation for Running Interactive Applications On Transient Servers

Abstract

Transient computing has become popular in public cloud environments for running delay-insensitive batch and data processing applications at low cost. Since transient cloud servers can be revoked at any time by the cloud provider, they are considered unsuitable for running interactive application such as web services. In this paper, we present VM deflation as an alternative mechanism to server preemption for reclaiming resources from transient cloud servers under resource pressure. Using real traces from top-tier cloud providers, we show the feasibility of using VM deflation as a resource reclamation mechanism for interactive applications in public clouds. We show how current hypervisor mechanisms can be used to implement VM deflation and present cluster deflation policies for resource management of transient and on-demand cloud VMs. Experimental evaluation of our deflation system on a Linux cluster shows that microservice-based applications can be deflated by up to 50% with negligible performance overhead. Our clusterlevel deflation policies allow overcommitment levels as high as 50%, with less than a 1% decrease in application throughput, and can enable cloud platforms to increase revenue by 30%.Before presenting our deflation techniques, we examine the efficacy and feasibility of deflating public cloud applications. We use publicly-available resource usage traces from two top-tier cloud providers, Azure [14] and Alibaba [15]. The goal of our analysis is to understand the feasibility of deflating CPU, memory, disk, and network allocations of real cloud applications, and specifically interactive web applications, under time-varying workloads that they exhibit. We seek to answer two key research questions through our feasibility analysis: (1) How much slack is present in cloud VMs and by how much can these VMs be safely deflated without any performance impact? (2) How does workload class and VM size impact the deflatability of VMs? Application Behavior under DeflationWe first present an abstract model to capture the performance behavior of an application under different amounts of resource Performance Slack Knee Linear Deflation %

Memory-harvesting VMs in cloud platforms

“…Summary. Prior work [4,14,43] did not consider the three major issues for production deployment of memory harvesting in real clouds: (1) The impact of memory harvesting on VM creation performance, (2) host memory fragmentation, and (3) NUMA spanning. Our work develops new techniques to address these challenges and improves existing techniques, such as the speed of runtime memory resizing, and develop new ones to mitigate these issues.…”

“…Second-level paging by the hypervisor to swap space or a remote server could be done to reclaim memory. [14] does this when cooperative memory trading cannot happen. However, cooperative memory trading on the same node involves less performance jitter (no I/O interrupts and no second-level page faults) and is much more efficient than paging to swap space or a remote node.…”

“…Though the second and third challenges may also imply performance issues for the workloads running on MHVMs, they are not as serious because these workloads should be inherently amenable to performance variability (including evictions). The prior work that studied memory harvesting in virtualized environments [14,43] considered none of these challenges and performance implications.…”

ReSQoV: A Scalable Resource Allocation Model for QoS-Satisfied Cloud Services

“…The simulator CloudSim Plus [7,21] is used to present and simulate the scenarios to assess the proposed algorithm and methodology. For the simulation, a virtual host with a Virtual Machine setup matching the instances of Amazon T2.medium is used.…”

LaSS

“…Resource deflation is a recently proposed approach for reclaiming resources by reducing the CPU and memory allocation of a virtual machine. The deflation approach proposed in [26,44] is based on the observation that virtual machines have slack in their resource allocation (since their allocated resources may not be 100% utilized by the application code executing inside them) and reclaiming a fraction of the VM's allocated resource can be done without a proportionate degradation in application performance. The deflation study in [26] has analyzed millions of production VMs in the Azure cloud and showed that typical slack can be up to 50%.…”