Availability-aware Service Placement Policy in Fog Computing Based on Graph Partitions

TL;DR

The paper tackles availability-aware service placement in fog computing under device and link failures and strict application deadlines. It introduces a two-phase policy that first assigns applications to well-connected device communities and then places service transitive closures within those communities, using First-Fit Decreasing and a fitness-guided First-Fit strategy. The approach is novel in jointly leveraging graph-based device partitioning and closure-based service grouping, and it demonstrates higher QoS and availability than a comparable ILP optimizer in YAFS-based simulations with dynamic failures. The work offers a scalable, resilience-focused placement method for IoT-fog ecosystems and points to future extensions for additional metrics such as cost and migration overhead.

Abstract

This paper presents a policy for service placement of fog applications inspired on complex networks and graph theory. We propose a twofold partition process based on communities for the partition of the fog devices and based on transitive closures for the application services partition. The allocation of the services is performed sequentially by, firstly, mapping applications to device communities and, secondly, mapping service transitive closures to fog devices in the community. The underlying idea is to place as many inter-related services as possible in the most nearby devices to the users. The optimization objectives are the availability of the applications and the Quality of Service (QoS) of the system, measured as the number of requests that are executed before the application deadlines. We compared our solution with an Integer Linear Programming approach, and the simulation results showed that our proposal obtains higher QoS and availability when fails in the nodes are considered.

Figures (7)

• Figure 1: Fog computing architecture.
• Figure 2: Example of fog device communities.
• Figure 3: Example of service transitive closures.
• Figure 4: Evolution of the QoS with regard to the fail of fog devices, in terms of the number of requests which satisfy application deadlines ($|RT_{RQ^n_{US_a,APP_x}} < DL_{APP_x}|$) compared with the total number of requests ($|RQ^n_{US_a,APP_x}|$).
• Figure 5: Number of IoT devices that get services in regard with the number of failed fog devices ($availability(APP_x)$).