Avaliando os impactos do uso de variantes para os algoritmos LRU e LFU no contexto de sistemas Video sob demanda

Abstract

The CISCO report (2020) presents a forecast about the development of the Internet in general, considering that it will be necessary an evolution at all levels of Internet Service Providers networks (ISPs) in order to be able to supply the total demand. The report by Cullen Cantor (2020) shows that the video traffic is responsible for more than 50% of network traffic and with this information it is possible to understand that the growth and optimization of the video-on-demand (VoD) proxy server can become crucial. The algorithms used within the VoD proxy servers represent an important part of the performance of this server, as they generate an increase in the success rate of these algorithms, an increase that may represent a decrease of traffic in parts of the network. Performing a simple modification in the Least Recently Used (LRU) and Least Frequently Used (LFU) algorithms can bring an increase in the success rate, such as the implementation of a fixed analysis time window, which brought an improvement around 20% in the LFU hit rate for scenarios with high Zipf number. This rate may vary depending on the scenario and the size of the window being used. The scenario used in this work was generated using the Poisson Distribution for the customer input interval and the Zipf Law that describes the frequency of the videos. A simulator developed specifically for the present work was used, which manages to simulate an environment in order to achieve a correct rate of scenarios without having all the memory and bandwidth of the simulation. Simulations were even performed for the Randon, LRU, LFU, First In First Out (FIFO), CC and CARTE algorithms in order to compare the results found.