Scalable distributed service migration via Complex Networks Analysis

TL;DR

This paper introduces cDSMA, a scalable distributed algorithm for optimal user-centric service placement using complex network analysis, demonstrating high accuracy and efficiency in diverse network topologies.

Contribution

It presents a novel metric based on complex network analysis and a scalable heuristic for distributed service migration, improving over local-search methods.

Findings

High accuracy and fast convergence of cDSMA

Robustness to network size and demand estimation errors

Outperforms local-search heuristics in simulations

Abstract

With social networking sites providing increasingly richer context, User-Centric Service (UCS) creation is expected to explode following a similar success path to User-Generated Content. One of the major challenges in this emerging highly user-centric networking paradigm is how to make these exploding in numbers yet, individually, of vanishing demand services available in a cost-effective manner. Of prime importance to the latter (and focus of this paper) is the determination of the optimal location for hosting a UCS. Taking into account the particular characteristics of UCS, we formulate the problem as a facility location problem and devise a distributed and highly scalable heuristic solution to it. Key to the proposed approach is the introduction of a novel metric drawing on Complex Network Analysis. Given a current location of UCS, this metric helps to a) identify a small subgraph of nodes with high capacity to act as service demand concentrators; b) project on them a reduced yet accurate view of the global demand distribution that preserves the key attraction forces on UCS; and, ultimately, c) pave the service migration path towards its optimal location in the network. The proposed iterative UCS migration algorithm, called cDSMA, is extensively evaluated over synthetic and real-world network topologies. Our results show that cDSMA achieves high accuracy, fast convergence, remarkable insensitivity to the size and diameter of the network and resilience to inaccurate estimates of demands for UCS across the network. It is also shown to clearly outperform local-search heuristics for service migration that constrain the subgraph to the immediate neighbourhood of the node currently hosting UCS.