Towards Deviceless Edge Computing: Challenges, Design Aspects, and Models for Serverless Paradigm at the Edge

Abstract

Recently, Cloud Computing, Edge Computing, and the Internet of Things (IoT) have been converging ever stronger, sparking creation of very large-scale, geographically distributed systems [1, 2]. Such systems intensively exploit Cloud Computing models and technologies, predominantly by utilizing large and remote data centers, but also nearby Cloudlets [3, 4] to enhance resource-constrained Edge devices (e.g., in terms of computation offloading [5-7] and data staging [8]) or to provide an execution environment for cloud-centric IoT/Edge applications [9, 10]. Serverless computing is an emerging paradigm, typically referring to a software architecture where application is decomposed into "triggers" and "actions" (or functions), and there is a platform that provides seamless hosting and execution of developer-defined functions (FaaS), making it easy to develop, manage, scale, and operate them. This complexity mitigation is mainly achieved by incorporating sophisticated runtime mechanisms into serverless or FaaS platforms. Hence, such platforms are usually characterized by fully automating many of the management and operations processes. Therefore, serverless computing can be considered as the next step in the evolution of Cloud platforms, such as PaaS, or more generally of the utility computing. While originally designed for centralized cloud deployments, the benefits of serverless paradigm become especially evident in the context of Edge Computing [11]. This is mainly because in such systems, traditional infrastructure and application management solutions are tedious, ineffective, error-prone, and ultimately very costly. Luckily, some of the existing serverless techniques, such

A Decentralized Framework for Serverless Edge Computing in the Internet of Things

“…With specific reference to edge computing environments, in [25] the authors propose an abstraction called Software-Defined Gateway, which hides the complexity of local IoT devices and offers a clean API for the integration with widely adopted serverless platforms; they focus on the DevOps aspects, and mention the issue of "scheduling [the] functions execution on loosely coupled and scarce edge resources" as an open research issue, which we address in this work. In [12] the authors take a more radical approach and assume that every edge node may decide at the same time where to route every incoming task (at a network-level) and which node has to execute the remaining processing, where applications are assumed to be modeled as directed computation graphs.…”

AuctionWhisk: Using an auction‐inspired approach for function placement in serverless fog platforms

“…1,10,12,122 Notably, the finely-grained execution units of serverless functions can lead to high load on central scheduling components, as Rausch et al 123 have shown in comparison to their earlier work. [124][125][126] The authors claim that there is no technical solution that is able to handle such scheduling of function execution across edge, intermediaries, and the cloud. On the other hand, precisely this flexible scheduling of function executions can maximize resource allocation and lead to better auction conditions for both providers as well as consumers.…”

“…Nevertheless, existing research has not taken into account the unique challenges and opportunities of fog‐based FaaS platforms, although it has been shown that the FaaS paradigm is a good fit for fog applications 1,10,12,122 . Notably, the finely‐grained execution units of serverless functions can lead to high load on central scheduling components, as Rausch et al 123 have shown in comparison to their earlier work 124‐126 . The authors claim that there is no technical solution that is able to handle such scheduling of function execution across edge, intermediaries, and the cloud.…”

Towards Auction-Based Function Placement in Serverless Fog Platforms

“…Nevertheless, existing research has not taken into account the unique challenges and opportunities of fog-based FaaS platforms, although it has been shown that the FaaS paradigm is a good fit for fog applications [103,104,122,20]. Notably, the finely-grained execution units of serverless functions can lead to high load on central scheduling components, as Rausch et al [108] have shown in comparison to their earlier work [90,89,109]. The authors claim that there is no technical solution that is able to handle such scheduling of function execution across edge, intermediaries, and the cloud.…”