Modeling Tradeoffs between mobility, cost, and performance in Edge Computing

TL;DR

This work addresses the problem of comparing edge versus centralized cloud deployments in mobility-rich networks by developing closed-form queuing models that capture mobility overheads and workload skews. It formulates mobility as a two-phase queueing process and derives latency bounds for edge as $M/M/1$ versus cloud as $M/M/k$, extending to $GI/G/k$ to account for non-Poisson arrivals. Validation covers experiments, simulations, and real VM traces, quantifying the extra edge capacity required via a Dynamic Traveling Repairman Problem (DTRP) based VM packing framework. The results provide actionable guidance for capacity planning, migration policies, and hybrid edge-cloud designs, illustrating how mobility and bursty workloads shape the edge's cost-performance tradeoffs.

Abstract

Edge computing provides a cloud-like architecture where small-scale resources are distributed near the network edge, enabling applications on resource-constrained devices to offload latency-critical computations to these resources. While some recent work showed that the resource constraints of the edge could result in higher end-to-end latency under medium to high utilization due to higher queuing delays, to the best of our knowledge, there has not been any work on modeling the trade-offs of deploying on edge versus cloud infrastructures in the presence of mobility. Understanding the costs and trade-offs of this architecture is important for network designers, as the architecture is now adopted to be part of 5G and beyond networks in the form of the Multi-access Edge Computing (MEC). In this paper we focus on quantifying and estimating the cost of edge computing. Using closed-form queuing models, we explore the cost-performance trade-offs in the presence of different systems dynamics. We model how workload mobility and workload variations influence these tradeoffs, and validate our results with realistic experiments and simulations. Finally, we discuss the practical implications for designing edge systems and developing algorithms for efficient resource and workload management.

Figures (8)

• Figure 1: Queuing models for distributed edge clouds versus centralized remote large-scale backend clouds.
• Figure 2: Two-phased single-server queue. Requests arrive and receive the first service ($\mu_1$). Then, with probability $r$, they enter the migration phase ($\mu_2$); otherwise, with probability $1 - r$, they exit the queue.
• Figure 3: Response times with mobility and migration overhead for different migration probabilities compared with our $(M/M/k)$ model. ($\mu_1 = \mu_2= 50$)
• Figure 4: Migration of a 128 MB context
• Figure 5: Simulation waiting times with sinusoidal arrivals compared with Equation \ref{['eq:wedgesin']} and \ref{['excesswait']}.

Theorems & Definitions (17)

• Lemma 3.1
• Corollary 3.2
• Corollary 3.3
• Corollary 3.4
• Lemma 3.5
• Corollary 3.6
• Lemma 4.1
• proof
• Lemma 4.2
• Lemma 4.3