To Offload or Not? An Analysis of Big Data Offloading Strategies from Edge to Cloud

Abstract

Major research efforts have been recently made to develop resource orchestration solutions to flexibly link edge nodes with centralised cloud resources so as to maximise the efficiency with which such a continuum of resources can be accessed by users. In this context, we consider the case of Big Data analytics in which total task completion time reductions can be achieved by routing tasks initially to edge servers and subsequently to cloud resources. We demonstrate that the task complexity of the computational jobs, the Wide Area Network (WAN) speed and the potential overload of edge servers (as reflected by CPU workloads) are crucial for achieving total task completion time reductions by offloading from edge to cloud resources. The edge-cloud orchestrators are situated in the edge nodes and, therefore, require continuous access to the parameters of WAN speeds (and their fluctuations), edge server CPU workload and the task complexities in Big Data analytics requirements to make accurate edge-to-cloud offloading decisions. With favourable values for these three parameters, large reductions in completion times can result from transfer of large-scale data from edge nodes to cloud resources, which can reduce the completion times by up to 97% and meet client deadlines for computational tasks with responsive and agile solutions.

Edge AI: A survey

“…Cisco introduced applications use cases of FC that include Fog Computing and the Internet of Everything (IoE), Video analytics optimisation. 4 Another work [115] discussed five likely areas for Fog Computing deployment: healthcare, smart grids, smart vehicles, urgent computing and augmented reality [116].…”

“…However, this technology faces key issues: security, speed of services and slow connections, which are often combined as low bandwidth/high latency and jitter as mobile devices offload computational and processing capacity to cloud computing services [3,4]. These challenges have been exacerbated by the continued proliferation of mobile and fixed Internet-connected devices [5].…”

“…OpenFog: https://opcfoundation.org/markets-collaboration/openfog/4 IoT, from Cloud to Fog Computing: https://blogs.cisco.com/perspectives/io t-from-cloud-to-fog-computing…”

Edge-Cloud Synergy: Unleashing the Potential of Parallel Processing for Big Data Analytics

“…The processing times are directly proportional to the data size (in GB) and inversely proportional to the server processing speed and to the computational complexity (in bits per instruction) of the task [24]. The data transfer times assume a constant WLAN speed but variable WAN speeds and the total time when utilising cloud resources is critically dependant on the WAN speed: high WAN speeds favour edge-to-cloud transfer while low WAN speeds favour edge node processing [14].…”

“…Recently, we have demonstrated the great reductions in task processing times if Big Data analytics can be flexibly moved from edge nodes to cloud resources if the combination of task complexity, processing powers, data transfer rates and edge node congestion are recognised [14]. In this paper, we explore how the parallel processing abilities of edge nodes and cloud servers can be combined to optimise computing performance, especially when data transfer rates are major constraints.…”