Making automatic, optimal workload placements in a multi-cloud, hybrid-cloud information technology (IT) infrastructure is an important problem that concerns DevOps. Here is one specific solution to the use case of social networking service provisioning, characterized by unpredictable peaks and lows, or burst and slump workloads.
Modern web service architecture workloads use the public cloud infrastructure to reduce costs. However, as business picks up, capital is available to the enterprise to own infrastructure. Optimizing the use of capex (own private cloud) versus opex (public cloud) for cost economics and performance remains a complex problem to solve. What makes the problem even more complex is the dynamic nature of pricing public cloud services. This paper proposes a least cost per connection algorithm as one solution approach to this challenge.
The paper presents a system model comprising job scheduling, resource allocation, load balancing in the cloud, and virtual machine load balancing. The algorithm takes the following parameters into consideration: cloud capability measurement, job scheduling, and scaling. The paper also presents results from experimentation. The authors analyze the impact of deploying the workload on an increasing number of cloud services. The paper finds the optimum point to be at four public clouds (multiple cloud use case).
The authors have defined the problem with a relatively narrow focus and assumptions. In a practical scenario, the most important issue to tackle in such configurations is the design concerning data movement. The authors have completely ignored the problem of making data available to the workload as the infrastructure is switched. Unless smart mechanisms are used to avoid data movement, the time and cost to move data can inhibit the applicability of the proposed solution.
The readability of the paper could have been improved by editors proficient in English technical writing.