Source-linked AI summary
Age and Value of Information: Non-linear Age Case
Antzela Kosta, Nikolaos Pappas, Anthony Ephremides, Vangelis Angelakis
TL;DR
Real-time status systems need metrics that capture the cost of stale information, especially when source samples are correlated. The paper introduces source-dependent CoUD and VoIU, analyzes them in an M/M/1 model, and finds that cost-function shape determines the CoUD–VoIU tradeoff.
Problem
Existing freshness measures based on elapsed age do not represent how source statistics affect the cost and predictive usefulness of stale information.
Method
The paper introduces source-dependent CoUD functions and VoIU, then analyzes their average behavior in an M/M/1 FCFS status-update system.
Results
CoUD is minimized near server utilization ρ ≈0.5 for the evaluated functions, while convex and concave costs create a CoUD–VoIU tradeoff and linear cost leaves VoIU independent of per-time-unit cost.
Takeaways & Limitations
The utilization objective can be selected according to whether an application prioritizes minimum average CoUD or maximum average VoIU.
Abstract
from arXiv · showhide
We consider a real-time status update system consisting of a source-destination network. A stochastic process is observed at the source, and samples, so called status updates, are extracted at random time instances, and delivered to the destination. In this paper, we expand the concept of information ageing by introducing the Cost of Update Delay (CoUD) metric to characterize the cost of having stale information at the destination. We introduce the Value of Information of Update (VoIU) metric that captures the reduction of CoUD upon reception of an update. The importance of the VoIU metric lies on its tractability which enables the minimization of the average CoUD.
I. INTRODUCTION
The paper extends freshness analysis from elapsed-time age to source-dependent staleness costs and update value. It motivates CoUD and VoIU for correlated processes and examines their tractability and use in M/M/1 systems.
- Real-time status updates carry generation timestamps and track time-varying content from sensing, actuation, and other applications.
- Age of Information measures elapsed time since the last received update was generated, with smaller average age indicating fresher information.
- Cost of Update Delay characterizes staleness through tunable cost functions, with tractable average-cost expressions derived for an M/M/1 FCFS system.
- When process samples are correlated, smaller information age can improve prediction accuracy; for independent samples, age primarily indicates delay.
- Value of Information of Update captures the reduction in destination uncertainty or CoUD produced by receiving an update.
- The paper obtains average VoIU for M/M/1 and relates VoIU-based server utilization to average-CoUD analysis.
II. SYSTEM MODEL
The system models timestamped updates traversing an FCFS queue, with staleness represented by a source-dependent cost function. VoIU measures the fractional benefit of each reception through the reduction in CoUD.
- Status updates are generated by a source, queued, and transmitted to a remote destination under FCFS service.
- Cost of Update Delay is a non-negative, monotonically increasing function of time between updates, chosen to reflect source data characteristics.
- Age is a special CoUD case measured in time units, whereas CoUD may use any payment function and relevant unit.
- The interarrival time Y_i measures time between successive update generations, while system time T_i includes queue waiting and network delay.
- VoIU is a bounded fraction in [0, 1], where 0 denotes minimum update benefit and 1 denotes maximum benefit.
- CoUD functions are matched to source autocorrelation: exponential for small correlation, logarithmic for large correlation, and linear for intermediate values.
A. Value of Information of Update Analysis
The analysis defines VoIU through the reduction in CoUD caused by receiving an update, considering linear, exponential, and logarithmic cost functions. It derives limiting behavior and a normalized time-average formulation for evaluating VoIU.
- Linear case: For fs(t) = αt, the general VoIU derivation uses the linear cost function and converges through sample averaging under ergodicity.The analysis also defines an effective arrival rate and relies on the expectation operator in its final expression.
- Linear case: For α = 1, CoUD becomes the timeliness of each arriving status update, recovering Age of Information.The corresponding cost reductions equal the interarrival times in the linear case.
- Exponential case: limYi→+∞VE,i = 1 and limTi→+∞VE,i = 1 − e^−αYi for the exponential cost function.These limits quantify the update value when interarrival time or system time becomes large.
- Logarithmic case: For the logarithmic case, limYi→+∞VL,i = 1 and limTi→+∞VL,i = 0.Large interarrival times maximize update value, whereas large system times minimize it because the update is less timely.
- Time-average VoIU: The normalized time-average VoIU is defined over an observation interval and evaluated separately for the three cost functions.The observation begins with an empty queue and ends after service completion of the samples arriving by time T.
III. COST OF UPDATE DELAY COMPUTATION FOR THE M/M/1 SYSTEM
The paper computes average CoUD for linear, exponential, and logarithmic cost functions in an M/M/1 status-update system with FCFS service. The derivation decomposes the time-average cost into inter-update regions and then obtains case-specific expressions, with numerical evaluation used for part of the logarithmic case.
- General computation: The average CoUD is derived by decomposing the time-average area into disjoint regions Q1, Qi for i ≥ 2, and a final area of width Tn.The derivation notes that Y and T are dependent, which complicates the general average-cost calculation.
- Case-specific derivations: The paper derives average CoUD for all three fs(t) cases and finds the optimum server policy for each case.For the linear function, each Qi region is represented as a trapezoid formed by the difference of two triangles.
- Case-specific derivations: The linear and exponential cases are evaluated analytically within the M/M/1 computation, including the exponential-case terms used to obtain average CoUD.The resulting average CoUD expression is reported after applying the relevant expressions.
- Case-specific derivations: For the logarithmic case, some terms are computed numerically and derivation details are omitted due to space limitations.This is the stated limitation of the logarithmic-case computation.
IV. VALUE OF INFORMATION UPDATE COMPUTATION FOR THE M/M/1 SYSTEM
The paper computes average VoIU for an M/M/1 queue with FCFS discipline by conditioning on interarrival times and applying iterated expectation. The linear case uses an analytic integral, while the exponential and logarithmic cases are evaluated numerically.
- Average VoIU computation: Average VoIU is computed for the M/M/1 queue with a first-come-first-served discipline using the same general procedure as the CoUD analysis.The analysis begins with the linear cost case and conditions the expected value on an interarrival time X = x.
- Linear case: For fs(t) = αt, the conditional expectation E[V] given X = x is combined with iterated expectation and the probability density function of X.The resulting integral is evaluated using a hypergeometric function.
- Non-linear cases: For fs(t) = e^αt − 1 and fs(t) = log(αt + 1), the expected values are computed numerically.The passage reports numerical evaluation for both non-linear cases without giving the omitted calculations.
V. NUMERICAL RESULTS
The M/M/1 numerical results compare average CoUD and VoIU across server utilization and three age-cost functions. CoUD is minimized near ρ ≈0.5, while VoIU is maximized near ρ ≈0.6, with function curvature determining the CoUD–VoIU tradeoff.
- For the linear case, increasing α proportionally increases average CoUD but leaves average VoIU and the optimal policy unchanged.VoIU is independent of α because each unit of time receives the same cost.
- ρ ≈0.5 minimizes average CoUD for the linear, exponential, and logarithmic functions when α = 0.1.The three functions exhibit similar behavior around the optimum.
- CE > CP > CL across utilization values, with exponential, linear, and logarithmic functions producing respectively higher, intermediate, and lower average CoUD.Away from the optimum, exponential CoUD grows most sharply and logarithmic CoUD most smoothly.
- ρ ≈0.6 maximizes VoIU across the evaluated functions and parameter values.The numerical evaluation considers α values 0.1, 0.5, and 1.
- Exponential cost produces higher CoUD and VoIU than linear cost, whereas logarithmic cost produces lower CoUD and VoIU.This comparison represents a tradeoff between timeliness and timeliness combined with transmission resources such as bandwidth.
VI. CONCLUSION
The study introduces CoUD and VoIU for freshness and update importance in an M/M/1 source-destination network. The results show that convex and concave cost functions create a CoUD–VoIU tradeoff, while linearity affects only CoUD.
- CoUD characterizes freshness through three cost functions chosen in relation to the observed process's autocorrelation.
- VoIU measures the reduction of CoUD and therefore the reduction of uncertainty after an update.
- Utilization can be selected to satisfy either the minimum average CoUD objective or the maximum average VoIU objective.
- In linear CoUD, VoIU is independent of the assigned per-time-unit cost; exponential and logarithmic cases instead exhibit a CoUD–VoIU tradeoff.In the latter cases, smaller average CoUD corresponds to smaller average VoIU.