Elastic-RAN: Um modelo de elasticidade multinível com grão adaptativo para Cloud Radio Access Network

Abstract

It is expected that, by 2020, cell phone networks will have been increased 10 times their coverage area, with more than 50 billion connected devices, supporting 100 times more user equipment and increasing data rate capacity by 1000 times. This will lead to a massive increase in data traffic, fostering the development of 5G and making industry and scientific initiatives turn their efforts to meet this demand. In this scenario, Cloud Radio Access Networks (C-RANs) based researches, an architecture that consolidates base stations (BSs) to a cloud-centric point, are gaining momentum, changing the idea of fixed and limited resources, as it benefits from one of the key features of Cloud Computing: resource elasticity. One of the major challenges in C-RAN architecture lies in the high complexity of orchestrating all of these computational resources in order to perform the requests processing with high performance and the lowest possible infrastructure cost. Considering this context, the present dissertation seeks to develop the Elastic-RAN model, proposing a multilevel non-blocking elasticity concept, with automatic orchestration of resources through the coordination of BBU Pools and their BBUs, with an adaptive elastic grain mechanism. The multilevel non-blocking elasticity allows it control the level of BBU Pool (physical machine), given the high volume of traffic and the suggested maximum distance between antennas and pools, and the level of BBU (virtual machine), due to the high CPU processing and memory required for the requests, so as not to penalize the current processing. The adaptive elastic grain mechanism allows the provisioning and mapping of resources on demand and at runtime, considering the current use of resources, so that each elastic action is performed with a grain close to the current processing needs. The Elastic-RAN model was evaluated through experiments that simulated different load profiles, which are executed in an intensive CPU and network traffic application, exploiting the transfer of streamings and processing block decoding. As a result, it was possible to observe that Elastic-RAN may achieve gains ranging from 4 % to 26 %, in relation to execution costs, when compared to the traditional elasticity approach. In addition, it achieved better efficiency for all load profiles and reduced by 55 % the amount of elastic operations required. Also, given the non-elasticity approach, cost gains were even higher, going from 51 % to 70 %.