Age Dispersion and Higher-Order AoI in Status Update Systems

Technical Summary: Age Dispersion and Higher-Order AoI in Status Update Systems

Problem Statement
The Age of Information (AoI) has become the standard metric for quantifying the freshness of information in status update systems. However, the paper argues that AoI alone does not fully capture the temporal consistency of updates. In certain applications, such as remote industrial support using cloud-rendered augmented reality (AR), updates must not only be fresh (low AoI) but also closely spaced in time. Large gaps between consecutive updates can lead to visual instability and motion sickness in AR applications, even if the most recent update is fresh. To address this gap, the authors introduce the concept of age dispersion as a measure of temporal consistency.

Methodology and Definitions
The paper analyzes a single-source status update system modeled as an M/G/1/1 queue (Poisson arrivals, general service times, single server, no waiting room) under a probabilistically preemptive policy. In this policy, an arriving packet enters service if the server is idle; if the server is busy, the new packet preempts the one in service with probability $\theta$ and is otherwise discarded.

The authors define the following key metrics:

1. Age Dispersion ($D$): The difference between the ages of the two most recently received updates. Formally, if $t'_i$ is the delivery time of the $i$-th update and $t_i$ is its generation time, the age dispersion at time $\tau$ is $D(\tau) = t_{N(\tau)} - t_{N(\tau)-1}$, where $N(\tau)$ is the index of the most recent update.
2. $k$-th Order Age Dispersion ($D^{(k)}$): The difference between the age of the most recently received update and the $(k+1)$-th most recently received update.
3. $k$-th Order AoI ($\Delta^{(k)}$): The age of the $k$-th most recently received update.

The authors establish a fundamental relationship between these metrics, showing that the average $k$-th order AoI is the sum of the average standard AoI ($\Delta$) and the average $(k-1)$-th order age dispersion:
$$\Delta^{(k)} = \Delta + D^{(k-1)}$$

Key Contributions and Analytical Results
The paper provides closed-form expressions for these metrics in the M/G/1/1 system:

• Average Age Dispersion ($D$): The authors derive that the average age dispersion is equal to the average inter-departure time, $E[Y]$. For the M/G/1/1 system with probabilistic preemption, this is given by:
  $$D = \frac{L_S(\lambda\theta)(\theta - 1) + 1}{\lambda\theta L_S(\lambda\theta)}$$
  where $L_S$ is the Laplace transform of the service time distribution and $\lambda$ is the arrival rate.

  • Special Cases: For a fully preemptive policy ($\theta=1$), $D_P = 1/(\lambda L_S(\lambda))$. For a non-preemptive policy ($\theta \to 0$), $D_{NP} = E[S] + 1/\lambda$.

• Average $k$-th Order AoI ($\Delta^{(k)}$): By combining the standard AoI results from prior literature with the derived dispersion formula, the paper characterizes the average $k$-th order AoI as:
  $$\Delta^{(k)} = \Delta + (k-1) \frac{L_S(\lambda\theta)(\theta - 1) + 1}{\lambda\theta L_S(\lambda\theta)}$$

Numerical Results and Observations
The authors evaluate the weighted sum $\Delta + \alpha D$ (which corresponds to $\Delta^{(k)}$ where $\alpha = k-1$) using Gamma-distributed service times with shape parameter $\kappa$ and rate $\beta$. Key findings include:

1. Impact of Hazard Rate: The effectiveness of preemption depends on the hazard rate of the service time distribution.
   • When $\kappa \le 1$ (decreasing or constant hazard rate), increasing the preemption probability $\theta$ reduces the average $(k+1)$-th order AoI.
   • When $\kappa > 1$ (increasing hazard rate), increasing $\theta$ increases the average $(k+1)$-th order AoI. This is because preempting a packet that is likely to finish soon (due to the increasing hazard rate) is counterproductive.
2. Preemption vs. Non-Preemption:
   • For $\kappa < 1$, the preemptive policy yields lower age dispersion ($D_P < D_{NP}$).
   • For $\kappa = 1$ (exponential service), $D_P = D_{NP}$.
   • For $\kappa > 1$, the preemptive policy yields higher age dispersion ($D_P > D_{NP}$).
3. Optimal Arrival Rate: The optimal arrival rate $\lambda$ that minimizes the higher-order AoI varies depending on the preemption probability $\theta$ and the service time distribution. High arrival rates are beneficial only when preemption is low or the service distribution allows for it; otherwise, they lead to excessive preemption of packets near completion.

Significance and Claims
The paper claims that age dispersion provides a necessary complement to AoI for applications requiring temporal consistency. By defining and characterizing higher-order AoI through the lens of age dispersion, the authors demonstrate that minimizing higher-order AoI inherently ensures both fresh information (low standard AoI) and temporally consistent data delivery (low age dispersion).

The work establishes a theoretical framework for analyzing these metrics in M/G/1/1 systems. The authors note that while this paper focuses on the M/G/1/1 model, future work could extend these characterizations to other queueing models (e.g., M/M/1, M/G/1/2, multi-source systems) and explore state-selective preemption policies where decisions depend on the current system state rather than a fixed probability.