Publicação
A Novel Approach to Load Balancing in P2P Overlay Networks for Edge Systems
| Resumo: | Edge computing aims at addressing some limitations of cloud computing by bringing computation towards the edge of the system, i.e., closer to the client. There is a panoply of devices that can be integrated into future edge computing platforms, from local datacenters and ISP points of presence, to 5G towers, and even, multiple user devices like smartphones, laptops, and IoT devices. For all of these devices to communicate fruitfully, we need to build systems that enable the seamless interaction and cooperation among these diverse devices. However, creating and maintaining these systems is not trivial since there are numerous types of devices with different capacities. This resource heterogeneity has to be taken into account so that different types of machines contribute to the management of the distributed infrastructure differently, and the operation of the overall system becomes more efficient. In this work, we addressed the challenges identified above by exploring unstructured overlay networks, that have been shown to be possible to manage efficiently and in a fully decentralized way, while being highly robust to failures. To that end, we devised a solution that adapts the number of neighbors of each device (i.e., how many other devices that device knows) according to the capacity of that device and the distribution of capacities of the other devices in the network, as to ensure that the load is fairly distributed between them and, as a consequence, improve the operation of other services atop the unstructured overlay network, for instance, reducing the latencies experienced when broadcasting information. This solution can be easily integrated into most existing peer-to-peer distributed systems, requiring just a slight adaptation to their membership protocol. To show the correction and benefits of our proposal, we evaluated it by comparing it with state of the art decentralized solutions to manage unstructured overlay networks, combining both simulation (to observe the performance of the solution at large scale) and prototype deployments in realistic distributed infrastructures. |
|---|---|
| Autores principais: | Menino, Vítor Hugo |
| Assunto: | peer-to-peer systems edge computing unstructured overlay networks |
| Ano: | 2022 |
| País: | Portugal |
| Tipo de documento: | dissertação de mestrado |
| Tipo de acesso: | acesso aberto |
| Instituição associada: | Universidade Nova de Lisboa |
| Idioma: | inglês |
| Origem: | Repositório Institucional da UNL |
| Resumo: | Edge computing aims at addressing some limitations of cloud computing by bringing computation towards the edge of the system, i.e., closer to the client. There is a panoply of devices that can be integrated into future edge computing platforms, from local datacenters and ISP points of presence, to 5G towers, and even, multiple user devices like smartphones, laptops, and IoT devices. For all of these devices to communicate fruitfully, we need to build systems that enable the seamless interaction and cooperation among these diverse devices. However, creating and maintaining these systems is not trivial since there are numerous types of devices with different capacities. This resource heterogeneity has to be taken into account so that different types of machines contribute to the management of the distributed infrastructure differently, and the operation of the overall system becomes more efficient. In this work, we addressed the challenges identified above by exploring unstructured overlay networks, that have been shown to be possible to manage efficiently and in a fully decentralized way, while being highly robust to failures. To that end, we devised a solution that adapts the number of neighbors of each device (i.e., how many other devices that device knows) according to the capacity of that device and the distribution of capacities of the other devices in the network, as to ensure that the load is fairly distributed between them and, as a consequence, improve the operation of other services atop the unstructured overlay network, for instance, reducing the latencies experienced when broadcasting information. This solution can be easily integrated into most existing peer-to-peer distributed systems, requiring just a slight adaptation to their membership protocol. To show the correction and benefits of our proposal, we evaluated it by comparing it with state of the art decentralized solutions to manage unstructured overlay networks, combining both simulation (to observe the performance of the solution at large scale) and prototype deployments in realistic distributed infrastructures. |
|---|