Server cluster management for compute heavy process types
Topcu, Ozan (2024)
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:amk-2024120131658
https://urn.fi/URN:NBN:fi:amk-2024120131658
Tiivistelmä
The purpose of this study was to create a solution that would act as some sort of autoscaler for computationally demanding servers. Multiple use cases were considered, and the Monte Carlo prediction simulation was used to imitate a real-life service as a benchmark.
The focus was on certain servers and process attributes and their optimisation. These values were measured on actual servers and the benchmarking tool and used as a foundation in the algorithm for server cluster management.
The results satisfied the scope of the project as the final version of the algorithm was capable of logically and accurately managing server count and types in real-time with fluctuating request traffic. The program was monitored thoroughly and possible future additions like considering the usage of machine learning in traffic prediction were discussed.
The focus was on certain servers and process attributes and their optimisation. These values were measured on actual servers and the benchmarking tool and used as a foundation in the algorithm for server cluster management.
The results satisfied the scope of the project as the final version of the algorithm was capable of logically and accurately managing server count and types in real-time with fluctuating request traffic. The program was monitored thoroughly and possible future additions like considering the usage of machine learning in traffic prediction were discussed.