Performance Characterization Model For A Cloud Control Plane

The control plane of a cloud infrastructure is responsible for managing and orchestrating the cloud resources, including provisioning, metering, monitoring, and security. Cloud user experience largely depends on the performance, availability and reliability of the control plane hardware and software stack managed by a cloud controller. Given that cloud is a large-scale multi-tenant environment, the load on cloud controller can increase dramatically with thousands of requests. The performance of the control plane is measured mainly by the time behavior and resource utilization and is affected by several factors such as the scale of the environment, load on cloud controller and resource constraints. Huge delays or time-out failures in serving the requests can become a common scenario in a highly scaled environment leading to bad user experience. Distributed tracing and root causing of such performance incidents, characterizing across the multi-layered control plane stack is a complex activity that needs a systematic approach and usage of cohesive tools for diagnosis as well as proactive resolution. To address these requirements, we discuss a Performance Characterization Model (PCM) in this paper, as a workflow for fine grained data driven evaluation to detect optimization areas and resolve performance problems. This model lays out a systematic approach through different phases of use-case analysis, load generation, metric characterization, data collection, analysis, and insight generation, as well as predicting performance thresholds based on data. We also demonstrate a specific case study for characterizing the performance of user interface of Cloud using the methodologies and tools of PCM.

I. Introduction

Large scale Infrastructure-as-a-Service (IaaS) clouds have expanded their horizons to encompass different enduser applications and use cases. Cloud environments in general provide the required scalability for applications – ability to increase or decrease the resources based on demand. However, the central management system of the Cloud – Cloud Control Plane – also plays an important role in scaling up and managing the underlying cloud infrastructure resources. The IaaS cloud control plane provides management and orchestration tools and consists of a set of services and components that are responsible for managing and controlling cloud resources. Any cloud resource request for provisioning, modifying, deleting, or observing the cloud resources must pass through the control plane. In a typical control plane architecture, at the bottom of the control plane stack is the hardware management layer that manages physical servers, storage, and network devices. Above that is the virtualization management layer which provides virtualization services to abstract the hardware resources and enable the creation of virtual machines. The next layer is the orchestration layer which typically consists of a resource manager, referred to as cloud controller from here on, which manages all the physical and virtual resources using the hardware management and virtualization layers. The top layer of the stack would be a front-end service integrated to the user interfaces providing self-provisioning capabilities including role-based access permissions, and workspaces for users. In a real-world cloud deployment, the specific responsibilities and services offered by a control plane can vary with different cloud providers like AWS or Azure [7], [8]. In general, the interfaces exposed to the users can be a web portal or UI, REST APIs and command line. The load exerted on this cloud control plane is mainly driven by user requests for observing and provisioning various cloud resources. Ensuring the availability, scalability and performance of this control plane is very critical for an IaaS cloud provider, to have a highly scalable cloud environment with seamless user experience enabling cloud users and requests to deploy and scale the resources required for their applications quickly and efficiently.