Dynamic adjustment of a MLFQ flow scheduler to improve cloud applications performance

State-of-the-art solutions for flow scheduling propose the use of Multi Level Feedback Queue (MLFQ) as a mechanism to avoid the requirement of prior information (i.e. agnosticism) regarding flow sizes. This is an important aspect to achieve the performance goals of high responsiveness and high throughput that is expected in Cloud Applications (e.g. search engines, social networks, and e-commerce sites). These goals are tightly associated with the prioritization of short flows (a few KB in size), the majority for these applications rather than long flows (several MB in size). However, these applications usually cannot provide information in advance about the size of the flows. In this paper, we analyze the feasibility of providing dynamic adjustment for a MLFQ-based scheduling system in such a way that it adapts itself to the time and space variations exhibited by Data Center Network (DCN) traffic without requiring prior information about workload properties.