Scalability and Nonlinear Performance Tuning in Storage Servers Academic Article uri icon

abstract

  • The ability to scale in order to handle an increased amount of work is critical to both a business and its Information Technology. However, there are diminishing returns as the number of scalability enhancements are increased. This article submits the fact that there is a need to consider performance tuning as opposed to only horizontal or vertical scalability. Optimization techniques based on a greedy algorithm are reviewed. For example, Constraint Programming, Mixed Integer Programming, and Local Search. The article further indicates that the complexities of performance tuning do not favor modelling it as a linear process. It proposes the use of evidence-based, probabilistic reasoning for system administrators to identify which parameters need to be tuned and how to tune them. An influence diagram is used as an example to show the complexities involved when tuning one parameter has a high probability of affecting many other parameters in the system.

keywords

  • Maria DB, Influence Diagrams, Probabilistic Reasoning, Decision-Making, Optimization, Greedy algorithms, Automation, Distributed databases

volume

  • 5

issue

  • 9