[Paper Review] A Temporal Approach to Stochastic Network Calculus

Jing Xie, Yuming Jiang+1 more|arXiv (Cornell University)|Dec 13, 2011

Network Traffic and Congestion Control26 references3 citations

TL;DR

This paper introduces a temporal approach to stochastic network calculus by modeling traffic and service in terms of cumulative inter-arrival and service times, enabling analysis of error-prone and contention-based networks like wireless and multi-access systems. The key contribution is a new time-domain framework that establishes delay and backlog bounds, superposition, and concatenation properties, linking them to existing space-domain models via model transformations.

ABSTRACT

Stochastic network calculus is a newly developed theory for stochastic service guarantee analysis of computer networks. In the current stochastic network calculus literature, its fundamental models are based on the cumulative amount of traffic or cumulative amount of service. However, there are network scenarios where direct application of such models is difficult. This paper presents a temporal approach to stochastic network calculus. The key idea is to develop models and derive results from the time perspective. Particularly, we define traffic models and service models based on the cumulative packet inter-arrival time and the cumulative packet service time, respectively. Relations among these models as well as with the existing models in the literature are established. In addition, we prove the basic properties of the proposed models, such as delay bound and backlog bound, output characterization, concatenation property and superposition property. These results form a temporal stochastic network calculus and compliment the existing results.

Motivation & Objective

• To address limitations in existing stochastic network calculus for analyzing networks with probabilistic service, such as error-prone wireless links and contention-based access.
• To develop a time-domain modeling approach that captures the temporal behavior of packet arrivals and service from the perspective of inter-arrival and service times.
• To establish fundamental properties—delay bound, backlog bound, output characterization, concatenation, and superposition—within the new temporal framework.
• To bridge the gap between time-domain and space-domain stochastic network calculus by deriving transformation rules between temporal and existing space-domain models.
• To enable performance analysis of complex network scenarios, such as superposition of Poisson processes, using the new temporal models.

Proposed method

• Define traffic models based on cumulative inter-arrival time between packets, denoted as $\Gamma(m,n) = a(n) - a(m)$, forming the basis of time-domain arrival curves.
• Define service models based on cumulative service time per packet, using the service time $\delta_n$ to characterize the temporal service process.
• Introduce two types of time-domain stochastic arrival curves: $v.w.d$ (with bounding function $h(y)$) and $v.b.c$ (with inverse function $z^{-1}_i(x)$), enabling probabilistic delay and backlog bounds.
• Establish model transformations between time-domain $v.w.d$ and space-domain $v.b.c$ stochastic arrival curves via inverse functions and suprema over time shifts.
• Prove the superposition property for $N$ flows by deriving the aggregate time-domain arrival curve $\lambda(n) = \inf\{\tau : \sum_{i=1}^N \alpha_i(\tau) \geq n\}$ and bounding function $h(y) = f(z^{-1}(y))$.
• Use min-plus algebra and max-plus algebra to formalize the temporal network calculus, ensuring consistency with established network calculus principles.

Experimental results

Research questions

• RQ1 How can stochastic network calculus be restructured from a time-domain perspective to better model probabilistic service in wireless and multi-access networks?
• RQ2 What are the fundamental properties—delay bound, backlog bound, output characterization, concatenation, and superposition—when traffic and service are modeled in the time domain?
• RQ3 How can time-domain stochastic arrival curves be transformed into their equivalent space-domain counterparts, and vice versa?
• RQ4 Can the superposition of multiple stochastic arrival processes, such as Poisson processes, be effectively characterized in the time domain?
• RQ5 What is the role of model transformation in enabling flexible and accurate performance analysis across different network scenarios?

Key findings

• The paper establishes a complete temporal stochastic network calculus framework based on cumulative inter-arrival and service times, enabling performance analysis in networks with probabilistic service.
• The time-domain $v.w.d$ stochastic arrival curve for the superposition of two independent Poisson processes is derived as $\lambda_s(n) = T_s \cdot n$ with $T_s < \frac{1}{\mu_1 + \mu_2}$ and bounding function $h_s(x) = 1 - (1 - \rho_s) \sum_{i=0}^{\lfloor x/T_s \rfloor} e^{-(\mu_1 + \mu_2)(iT_s - x)} \frac{[(\mu_1 + \mu_2)(iT_s - x)]^i}{i!}$.
• The superposition property is proven for time-domain $v.w.d$ arrival curves, showing that the aggregate flow has a $v.w.d$ SAC with bounding function $h(y) = f(z^{-1}(y))$, where $f = f_1 \otimes \cdots \otimes f_N$.
• A transformation method is established between time-domain $v.w.d$ and space-domain $v.b.c$ stochastic arrival curves, enabling cross-framework analysis and validation.
• The concatenation and backlog bound properties are proven in the time domain, demonstrating that the new framework supports core network calculus operations.
• The framework is shown to be applicable to challenging scenarios such as wireless networks with retransmissions and multi-access networks with random backoff, where space-domain models are difficult to apply directly.