Vertelas - Automated User-Controlled Vertical Elasticity in Existing Commercial Clouds

Cloud computing has become ubiquitous as a way for users to obtain scalable services. Commercial Infrastructure-as-a-Service (IaaS) providers offer pay-as-you-go Virtual Machines (VM) to users to run their applications. One of the main benefits of using IaaS is the ability to increase or decrease the size of a VM (amount of RAM, cpu, etc.) on demand, based on requirement. This approach is known as vertical elasticity, and results in reduced costs over the lifetime of an application for the user. However, most IaaS providers including Amazon and Google, do not offer vertical elasticity out-of-the-box. Instead, they offer horizontal elasticity, which is the ability to vary the number of VMs based on requirement. In a myriad of scenarios such as non-parallelizable applications and scientific computations, vertical elasticity instead of horizontal elasticity is a must to reduce costs. This paper proposes a framework called Vertelas, that users can utilize within their current cloud based VMs to achieve vertical elasticity and save costs. Vertelas automatically predicts the required resources, and moves the running application to smaller or larger VMs. By utilizing Linux container migrations, no modifications or support on the part of the service providers is required, making the approach provider-agnostic. Experimental evaluations on a real world IaaS provider (AWS), with real world load traces, demonstrate the substantial reductions in costs as opposed to running the application inside a single, fixed-size VM.