Palumbo, Fabio (2021) Cloud and mobile infrastructure monitoring for latency and bandwidth sensitive applications. [Tesi di dottorato]

Preview

Text Palumbo_Fabio_33.pdf Download (9MB) | Preview

Collection description

This PhD thesis involves the study of cloud computing infrastructures (from the networking perspective) to assess the feasibility of applications gaining increasing popularity over recent years, including multimedia and telemedicine applications, demanding low, bounded latency and sufficient bandwidth. I also focus on the case of telemedicine, where remote imaging applications (for example, telepathology or telesurgery) need to achieve a low and stable latency for the remote transmission of images, and also for the remote control of such equipment. Another important use case for telemedicine is denoted as remote computation, which involves the offloading of image processing to help diagnosis; also in this case, bandwidth and latency requirements should be enforced to ensure timely results, although they are less strict compared to the previous scenario. Nowadays, the capability of gaining access to IT resources in a rapid and on-demand fashion, according to a pay-as-you-go model, has made the cloud computing a key-enabler for innovative multimedia and telemedicine services. However, the partial obscurity of cloud performance, and also security concerns are still hindering the adoption of cloud infrastructure. To ensure that the requirements of applications running on the cloud are satisfied, there is the need to design and evaluate proper methodologies, according to the metric of interest. Moreover, some kinds of applications have specific requirements that cannot be satisfied by the current cloud infrastructure. In particular, since the cloud computing involves communication to remote servers, two problems arise: firstly, the core network infrastructure can be overloaded, considering the massive amount of data that has to flow through it to allow clients to reach the datacenters; secondly, the latency resulting from this remote interaction between clients and servers is increased. For these, and many other cases also beyond the field of telemedicine, the Edge and Fog computing paradigms were introduced. In these new paradigms, the IT resources are deployed not only in the core cloud datacenters, but also at the edge of the network, either in the telecom operator access network or even leveraging other users' devices. The proximity of resources to end-users allows to alleviate the burden on the core network and at the same time to reduce latency towards users. Indeed, the latency from users to remote cloud datacenters encompasses delays from the access and core networks, as well as the intra-datacenter delay. Therefore, this latency is expected to be higher than that required to interconnect users to edge servers, which in the envisioned paradigm are deployed in the access network, that is, nearby final users. Therefore, the edge latency is expected to be reduced to only a portion of the overall cloud delay. Moreover, the edge and central resources can be used in conjunction, and therefore attention to core cloud monitoring is of capital importance even when edge architectures will have a widespread adoption, which is not the case yet. While a lot of research work has been presented for monitoring several network-related metrics, such as bandwidth, latency, jitter and packet loss, less attention was given to the monitoring of latency in cloud and edge cloud infrastructures. In detail, while some works target cloud-latency monitoring, the evaluation is lacking a fine-grained analysis of latency considering spatial and temporal trends. Furthermore, the widespread adoption of mobile devices, and the Internet of Things paradigm further accelerate the shift towards the cloud paradigm for the additional benefits it can provide in this context, allowing energy savings and augmenting the computation capabilities of these devices, creating a new scenario denoted as mobile cloud. This scenario poses additional challenges for its bandwidth constraints, accentuating the need for tailored methodologies that can ensure that the crucial requirements of the aforementioned applications can be met by the current infrastructure. In this sense, there is still a gap of works monitoring bandwidth-related metrics in mobile networks, especially when performing in-the-wild assessment targeting actual mobile networks and operators. Moreover, even the few works testing real scenarios typically consider only one provider in one country for a limited period of time, lacking an in-depth assessment of bandwidth variability over space and time. In this thesis, I therefore consider monitoring methodologies for challenging scenarios, focusing on latency perceived by customers of public cloud providers, and bandwidth in mobile broadband networks. Indeed, as described, achieving low latency is a critical requirement for core cloud infrastructures, while providing enough bandwidth is still challenging in mobile networks compared to wired settings, even with the adoption of 4G mobile broadband networks, expecting to overcome this issue only with the widespread availability of 5G connections (with half of total traffic expected to come from 5G networks by 2026). Therefore, in the research activities carried on during my PhD, I focused on monitoring latency and bandwidth on cloud and mobile infrastructures, assessing to which extent the current public cloud infrastructure and mobile network make multimedia and telemedicine applications (as well as others having similar requirements) feasible.

Downloads

Actions (login required)

View Item