Application Placement Approaches to Improve Quality of Service in Fog Computing

Abstract

Fog Computing (FC) addresses Cloud Computing's (CC) limitations by utilizing distributed computational devices, known as fog devices, near the Internet of Things (IoT) environment to support a wide range of IoT applications. In FC, to ensure Quality of Service (QoS), users need to specify a placement plan for distributing IoT applications among fog devices for processing; this is known as the application placement problem (APP). With a potentially huge number of fog devices and applications, solving the APP can be decentralized, i.e., independent optimization can be performed in parallel for different clusters of fog devices, thus mitigating the networking and computing overhead and enhancing the QoS consequently. In this approach, clusters lacking sufficient fog devices may propagate undeployed applications to other clusters, potentially leading to uncertain fulfillment of QoS constraints, i.e., delay bounds on response time. Moreover, deploying applications based on available resources at the placement decision time might result in an increased number of propagated applications among clusters. Additionally, the heterogeneity in fog devices' capabilities and the variations in IoT application characteristics, such as computing and networking intensity, and delay sensitivity, pose challenges in choosing competent applications for powerful fog devices. Assigning specific applications to these powerful devices may result in delay violations for other applications on less powerful ones, potentially leading to propagating the latter ones to other clusters. A raise in the number of propagated applications, especially those with data streams, might lead to increased networking congestion, resulting in extended response time and potential violation in delays, particularly for delay-sensitive applications.

This thesis proposes three approaches aiming to improve the QoS of IoT applications, i.e., delays. First, an improved application placement approach through parallel collaboration (ParColl) is proposed to increase the probability of placing propagated applications within their delays, incorporating algorithms to enable parallel searching and manage the searching process. Second, an improved application placement approach through postponement (PostP) is proposed to maximize the number of non-propagated applications meeting their delays, employing algorithms to postpone placement of undeployed applications, instead of propagating them, if such postponement ensures their delays. Third, an application placement approach maximizing response times for applications while meeting delays through cluster-wide resource selection CWRS) is proposed. CWRS ensures that powerful fog devices are reserved for applications needing them to meet their delays, minimizing violations on other devices whenever possible. Experimental results of implementing the proposed approaches in iFogSim show an improvement in the percentage of applications processed within their QoS constraints and a reduction in average delay violation times compared to existing approaches.