Nowadays, the growing demand for large-scale infrastructure computing has been observed, as well as the technical and economic impacts/innovations that it brings with. Cloud Computing, Grid Computing, or other kinds of large-scale infrastructure’ providers, usually have deployed a complex system to manage its resources – of course envisaging sustainable profits in the long run. Therefore, providers need to fully understand the business impacts on managing their resources to properly adjust their management systems.
One key part of large-scale computing infrastructure’s management systems is the scheduling strategy that is adopted. The answer for questions like “In which time frame jobs will be executed?”, “Which physical machines will handle the execution?”, and “Allocating more cores to complete a certain job justify the effort?” usually influence and impact on costs. The answer for such questions is exactly what a scheduling strategy is about.
While the general effects of a scheduling strategy can be estimated in theory, such estimation can be somehow limited to the provider since usually it is designed to non-specific environments (e.g., for general and not fine-grained parameters). Moreover, it is complex and difficult to assess the specific cost impact of a scheduling strategy in a detailed and realistic manner.
This thesis aims to design and implement a prototypical decision support tool that helps providers to visualize job allocations according to the configurable set of input parameters that matters for resource scheduling. The implementation will be based on a data set from an existent large-scale computing infrastructure, and will help to analyze which kind of information that the decision support tool can highlight to better understand how a current scheduling strategy could be further tuned – aiming to save resources and, consequently, save costs.
Supervisors: Prof. Dr. Burkhard Stiller, Dr. Guilherme Sperb Machado
back to the main page