Automated Fine-Grained CPU Provisioning for Virtual Machines

Abstract

Ideally, the pay-as-you-go model of Infrastructure as a Service (IaaS) clouds should enable users to rent just enough resources (e.g., CPU or memory bandwidth) to fulfill their service level objectives (SLOs). Achieving this goal is hard on current IaaS offers, which require users to explicitly specify the amount of resources to reserve; this requirement is nontrivial for users, because estimating the amount of resources needed to attain application-level SLOs is often complex, especially when resources are virtualized and the service provider colocates virtual machines (VMs) on host nodes. For this reason, users who deploy VMs subject to SLOs are usually prone to overprovisioning resources, thus resulting in inflated business costs.This article tackles this issue with AutoPro: a runtime system that enhances IaaS clouds with automated and fine-grained resource provisioning based on performance SLOs. Our main contribution with AutoPro is filling the gap between application-level performance SLOs and allocation of a contended resource, without requiring explicit reservations from users. In this article, we focus on CPU bandwidth allocation to throughput-driven, compute-intensive multithreaded applications colocated on a multicore processor; we show that a theoretically sound, yet simple, control strategy can enable automated fine-grained allocation of this contended resource, without the need for offline profiling. Additionally, AutoPro helps service providers optimize infrastructure utilization by provisioning idle resources to best-effort workloads, so as to maximize node-level utilization.Our extensive experimental evaluation confirms that AutoPro is able to automatically determine and enforce allocations to meet performance SLOs while maximizing node-level utilization by supporting batch workloads on a best-effort basis.

vmBBProfiler: a black-box profiling approach to quantify sensitivity of virtual machines to shared cloud resources

“…Rao et al [7] proposed VCONF, an auto-configuration RL-based approach, to automatically adjust CPU and Memory shares of VMs to avoid performance degradation. Other approaches [8][9][10][11][12], including Q-cloud [8] and TRACON [10], control applications' response times inside VMs only through adjusting their CPU-shares. Bartolini et al [12] proposed AutoPro to take a user-defined metric and adjust VMs' resources to close the gap between their desired performances and their current ones.…”

“…Other approaches [8][9][10][11][12], including Q-cloud [8] and TRACON [10], control applications' response times inside VMs only through adjusting their CPU-shares. Bartolini et al [12] proposed AutoPro to take a user-defined metric and adjust VMs' resources to close the gap between their desired performances and their current ones. AutoPro uses a PI controller to asymptotically close this gap and can work with any metric (eg, frame/s) as long as developers can provide it.…”

“…vmDataCollector(vmName,startTime,finishTime) collects/records performance of a VM under imposed limitations through polling several Hypervisor level metrics (eg, CPU utilization): it neither demands nor needs any specific metric from the VM itself. This makes vmBBProfiler unique when compared with many other similar approaches that either ask developers to provide specific metrics to reflect a VM's performance [5,6,8,[11][12][13]21], or use peripheral systems to collect external metrics to reflect such performance [24]. Because, the vmBBProfiler remains fully agnostic to the actual internal processes of a VM, we refer to our approach as a 'Black-Box' technique.…”

Quality of Service Driven Runtime Resource Allocation in Reconfigurable HPC Architectures

“…Figure 2 depicts the high level structure of the throughputbased resource management algorithm. To achieve modularity, we follow an approach already in use for other resource management problems [12]: we separate the algorithm in two layers having a clear interface between them, the Job Controllers and the Resource Broker. The algorithms used in the job controllers and in the resource broker are based upon heuristics; nevertheless, the split in two levels makes it simple to exploit other techniques for the per-job controller (e.g., formal control theory) keeping low the complexity of the controllers, and leaving to the resource broker the task of coping with constraints.…”

Autonomic thread scaling library for QoS management

“…Different control techniques have been explored and a fairly comprehensive summary can be found in [12]. To attain a good control, multiple types of actions have been investigated such as CPU cores affinity management [9], DVFS control [10], task mapping [13], CPU time allocation [14], and cache coloring [15].…”