Browsing by Author "Chavula, Josiah"

Now showing 1 - 8 of 8
  • Thumbnail Image
    Item
    Open Access
    An analysis of internet traffic flow in SANReN using active and passive measurements
    (2021) Salie, Luqmaan; Chavula, Josiah
    National research and education networks (NRENs) in developing regions such as Africa experience various performance issues due to inadequate infrastructure and resources. The South African National Research Network (SANReN) connects universities, research institutions, and oversees science projects such as the Square Kilometre Array. In this study, we conduct active and passive measurements to assess the performance of SANReN and to identify problem areas in the network. Active measurements were done to determine network performance when accessing SANReN internally (using PerfSONAR) and externally (using Speedchecker). We found that SANReN needs to be reinforced in and around Port Elizabeth, Cape Town, and Durban. Universities in these cities had the highest delays and page load times. We found that the network traffic flowing from PE uses circuitous routes to flow to universities in Johannesburg and Pretoria, causing high delays (medians of 25.26 ms to WITS, 25.47 ms to UJ, and 25.95 ms to UNISA) and high page load times (medians of 237.07 ms to WITS, 272.09 ms to UNISA, 280.47 ms to UJ transferring 31594 bytes of data). Using Cape Town as the traffic source resulted in a low median throughput of 5.47 Gbps for internal active measurements. Throughput from Durban to Cape Town was low as well (4.91 Gbps), causing high page load times between these two cities (medians of 350.32 and 305.22 ms from Durban to UCT and UWC respectively). SANReN's passive measurements results show us that there is a ratio of 11.16:1 for download speed to upload speed. We also observe a ratio of 2.29:1 for outbound flows (uploads) to inbound flows. Thus, majority of traffic flows experience low throughput amounts. Based on the test results, we design an SDN model and compare its performance to SANReN. The SDN model's results show that it would increase throughput while decreasing delays and page load times.
  • No Thumbnail Available
    Item
    Open Access
    Automated machine learning driven quality of service management in resource-constrained software defined networks
    (2023) White, Keegan; Chavula, Josiah
    Community networks are a means to bridge the connectivity gaps present in low-income and rural areas. Many of these networks are resource-constrained, mesh-based, and connected to the Internet via low-capacity links. These characteristics result in poor network performance. Software Defined Networking facilitates dynamic resource allocation to address real-time network degradation. Using the Software Defined Networking paradigm, methods to identify what traffic to allocate resources to offer a promising solution to common network issues in community networks. This dissertation presents a novel end-toend framework that uses deep learning models to facilitate real-time resource allocation in a resource-constrained network based on heuristics for traffic prioritisation. The deep learning models utilised by the framework are trained on data gathered from a community network and extensively tested in online network simulations. The results of this study convey that deep learning enabled Software Defined Networks can improve network throughput and decrease packet loss in real-time, thus improving network Quality of Service.
  • Thumbnail Image
    Item
    Open Access
    Cloud performance efficiency in Africa
    (2022) Babasanmi, Opeoluwa Victor; Chavula, Josiah
    Internet and Public Cloud adoption has been growing all over the world. An increasing amount of research has been conducted on Internet Performance across different continents. Major Cloud Providers have a growing presence in Africa, but little study has been conducted on Cloud Performance in Africa. This study undertakes to determine network performance from Africa to Public Cloud providers and compare this with what is achievable in a more developed continent like Europe. To achieve this, RIPE Atlas platform is used to run latency and traceroute measurements from RIPE Atlas endpoints in Africa and Europe to Public Cloud CDN endpoints and Virtual servers in Datacenters publicly available in both continents. Reverse measurements are also conducted from the Virtual servers to non-RIPE endpoints in both Africa and Europe. We find that countries with high network latencies in Africa are using CDN endpoints outside of Africa and, in some cases, make use of circuitous routes to Cloud destinations in Africa. In Europe, we found this to be different, as majority of CDN endpoints used were local to the continent, thereby leading to better CDN performance. While we see that using less expensive CDN PoPs in Africa could provide up to 87 percent performance improvement over relying on the Cloud Regions, Europe achieved up to 142 percent improvement. Following the results of this study, we recommend that Cloud providers should continue to increase their CDN presence in Africa and work with local ISPs to improve routing to ensure that local Cloud infrastructure is optimized for network traffic within the continent.
  • Thumbnail Image
    Item
    Open Access
    Effect of content caching on user QoE in iNethi community network
    (2022) Mwenje, Chikomborero; Chavula, Josiah
    The aim of this research was to determine effectiveness of content caching in community networks. This was achieved by measuring network performance and user quality of experience. The network performance was measured by performing latency and throughput tests in the network. Latency, throughput and video performance measurements were carried out in the Ocean View community network between the main server and access points in 8 locations in the network. The same measurements were carried out in the simulated network using 3 different caching strategies. The network measurements showed that caches resulted in lower latency and higher throughput in the simulated networks. Also caches resulted in less time required for initial buffering to occur, less time taken for video to start playing on its best quality and less time for the video to complete playing compared to the main server. This suggests that content caching improved resource utilisation, network performance and user QoE. A comparison between the cache placement strategies was done to determine which strategy performed best in the simulated network. Latency, throughput and video performance measurements were carried out for geography, delay and hop count cache placement. From the results obtained, hop count cache placement resulted in the lowest average latency, highest average throughput and best video performance. The simulated network was then expanded and measurements were again carried out in the expanded network. The expanded network adopted hop count cache placement to determine if the network would continue to improve performance when more caches are added. Expansion of the network resulted in lower average latency, higher average throughput and better video performance at the caches compared to the main server. This reinforces the effectiveness of content caching in improving network performance even in larger networks.
  • Thumbnail Image
    Item
    Open Access
    Improving Pan-African research and education networks through traffic engineering: A LISP/SDN approach
    (2017) Chavula, Josiah; Suleman, Hussein; Densmore, Melissa
    The UbuntuNet Alliance, a consortium of National Research and Education Networks (NRENs) runs an exclusive data network for education and research in east and southern Africa. Despite a high degree of route redundancy in the Alliance's topology, a large portion of Internet traffic between the NRENs is circuitously routed through Europe. This thesis proposes a performance-based strategy for dynamic ranking of inter-NREN paths to reduce latencies. The thesis makes two contributions: firstly, mapping Africa's inter-NREN topology and quantifying the extent and impact of circuitous routing; and, secondly, a dynamic traffic engineering scheme based on Software Defined Networking (SDN), Locator/Identifier Separation Protocol (LISP) and Reinforcement Learning. To quantify the extent and impact of circuitous routing among Africa's NRENs, active topology discovery was conducted. Traceroute results showed that up to 75% of traffic from African sources to African NRENs went through inter-continental routes and experienced much higher latencies than that of traffic routed within Africa. An efficient mechanism for topology discovery was implemented by incorporating prior knowledge of overlapping paths to minimize redundancy during measurements. Evaluation of the network probing mechanism showed a 47% reduction in packets required to complete measurements. An interactive geospatial topology visualization tool was designed to evaluate how NREN stakeholders could identify routes between NRENs. Usability evaluation showed that users were able to identify routes with an accuracy level of 68%. NRENs are faced with at least three problems to optimize traffic engineering, namely: how to discover alternate end-to-end paths; how to measure and monitor performance of different paths; and how to reconfigure alternate end-to-end paths. This work designed and evaluated a traffic engineering mechanism for dynamic discovery and configuration of alternate inter-NREN paths using SDN, LISP and Reinforcement Learning. A LISP/SDN based traffic engineering mechanism was designed to enable NRENs to dynamically rank alternate gateways. Emulation-based evaluation of the mechanism showed that dynamic path ranking was able to achieve 20% lower latencies compared to the default static path selection. SDN and Reinforcement Learning were used to enable dynamic packet forwarding in a multipath environment, through hop-by-hop ranking of alternate links based on latency and available bandwidth. The solution achieved minimum latencies with significant increases in aggregate throughput compared to static single path packet forwarding. Overall, this thesis provides evidence that integration of LISP, SDN and Reinforcement Learning, as well as ranking and dynamic configuration of paths could help Africa's NRENs to minimise latencies and to achieve better throughputs.
  • No Thumbnail Available
    Item
    Open Access
    Internet transparency in developing African regions: case of the DR Congo
    (2023) Munganga, Dieudonne; Chavula, Josiah
    This thesis investigates internet transparency and connectivity in the Democratic Republic of Congo (DRC) using a combination of secondary data analysis, online surveys, and primary data collected through Internet measurement on personal devices. The research aims to understand the perception of internet performance by the users, the level of interconnection among Autonomous systems and the Quality of Service (QoS) and Quality of Experience (QoE) provided by broadband networks in the DRC, and how it compares to other Central African countries. The research found that users in the DRC have a low level of satisfaction with internet performance and that the country's internet infrastructure and Autonomous systems are not well interconnected, leading to poor network performance. Additionally, the research found that the QoS and QoE provided by broadband networks in the DRC are far from optimal and lower than other Central African countries. Furthermore, the research revealed that there is a lack of transparency in the DRC's internet structure, with certain networks exerting a significant level of influence on users' internet connectivity. These findings indicate a need for increased transparency and improved internet infrastructure in the DRC to better serve the needs of its citizens and support economic and social development. The thesis concludes with a section on future work, highlighting potential avenues for further research such as enhancing the Internet measurement aspect of the study, using machine learning techniques to analyze the data, conducting a study on the impact of internet access and quality on social development, and developing a real-time monitoring system for internet connectivity in the DRC.
  • Thumbnail Image
    Item
    Open Access
    Investigating optimal internet data collection in low resource networks
    (2023) Sharma, Taveesh; Chavula, Josiah
    Community networks have been proposed by many networking experts and researchers as a way to bridge the connectivity gaps in rural and remote areas of the world. Many community networks are built with low-capacity computing devices and low-capacity links. Such community networks are examples of low resource networks. The design and implementation of computer networks using limited hardware and software resources has been studied extensively in the past, but scheduling strategies for conducting measurements on these networks remains an important area to be explored. In this study, the design of a Quality of Service monitoring system is proposed, focusing on performance of scheduling of network measurement jobs in different topologies of a low-resource network. We also propose a virtual network testbed and perform evaluations of the system under varying measurement specifications. Our results show that the system is capable of completing almost 100% of the measurements that are launched by users. Additionally, we found that the error due to contention for network resources among measurements stays constant at approximately 34% with increasing number of measurement nodes.
  • Thumbnail Image
    Item
    Open Access
    Using deep learning to classify community network traffic
    (2022) Matowe, Chiratidzo; Chavula, Josiah
    Traffic classification is an important aspect of network management. This aspect improves the quality of service, traffic engineering, bandwidth management and internet security. Traffic classification methods continue to evolve due to the ever-changing dynamics of modern computer networks and the traffic they generate. Numerous studies on traffic classification make use of the Machine Learning (ML) and single Deep Learning (DL) models. ML classification models are effective to a certain degree. However, studies have shown they record low prediction and accuracy scores. In contrast, the proliferation of various deep learning techniques has recorded higher accuracy in traffic classification. The Deep Learning models have been successful in identifying encrypted network traffic. Furthermore, DL learns new features without the need to do much feature engineering compared to ML or Traditional methods. Traditional methods are inefficient in meeting the demands of ever-changing requirements of networks and network applications. Traditional methods are unfeasible and costly to maintain as they need constant updates to maintain their accuracy. In this study, we carry out a comparative analysis by adopting an ML model (Support Vector Machine) against the DL Models (Convolutional Neural Networks (CNN), Gated Recurrent Unit (GRU) and a hybrid model: CNNGRU to classify encrypted internet traffic collected from a community network. In this study, we performed a comparative analysis by adopting an ML model (Support vector machine). Machine against DL models (Convolutional Neural networks (CNN), Gated Recurrent Unit (GRU) and a hybrid model: CNNGRU) and to classify encrypted internet traffic that was collected from a community network. The results show that DL models tend to generalise better with the dataset in comparison to ML. Among the deep Learning models, the hybrid model outperformed all the other models in terms of accuracy score. However, the model that had the best accuracy rate was not necessarily the one that took the shortest time when it came to prediction speed considering that it was more complex. Support vector machines outperformed the deep learning models in terms of prediction speed.