Source-linked AI summary
On The Age Of Information In Status Update Systems With Packet Management
Maice Costa, Marian Codreanu, Anthony Ephremides
TL;DR
The paper asks how communication systems can keep information about an observed process timely when updates compete for limited transmission resources. It models the source-destination link with queues and packet management, including discarding or replacing stale packets, and analyzes average and peak age. Packet replacement generally gives the best reported age performance, while the preferred policy can depend on arrival rates and whether average or peak age is considered.
Problem
Communication networks need measures and mechanisms for delivering fresh process information because outdated messages may lose their value.
Method
The paper models the source-destination link as finite-capacity queues with packet management policies that discard or replace updates, and analyzes average and peak age.
Results
Packet replacement produces the smallest average age and the best peak-age results among the evaluated models, with approximately 5% lower average age at λ = 0.6 than the other models.
Takeaways & Limitations
Peak age provides a simpler freshness-oriented metric with a complete distribution and supports evaluating whether information stays below a specified threshold.
Abstract
from arXiv · showhide
We consider a communication system in which status updates arrive at a source node, and should be transmitted through a network to the intended destination node. The status updates are samples of a random process under observation, transmitted as packets, which also contain the time stamp to identify when the sample was generated. The age of the information available to the destination node is the time elapsed since the last received update was generated. In this paper, we model the source-destination link using queuing theory, and we assume that the time it takes to successfully transmit a packet to the destination is an exponentially distributed service time. We analyze the age of information in the case that the source node has the capability to manage the arriving samples, possibly discarding packets in order to avoid wasting network resources with the transmission of stale information. In addition to characterizing the average age, we propose a new metric, called peak age, which provides information about the maximum value of the age, achieved immediately before receiving an update.
I. INTRODUCTION
The paper studies timely delivery of sampled process information through queues, focusing on packet management and introducing peak age alongside average age.
- Motivation: Age of information is the elapsed time since the most recently received status update was generated.Packets contain process information and the sample-generation timestamp; age is defined as ∆(t) = t − U(t).
- Packet management: The source can discard arriving packets to avoid transmitting stale information, using three packet-management policies.The policies respectively discard packets finding the server busy, retain one waiting packet, or replace the waiting packet with a newer arrival.
- Scope: The paper notes that choosing packet management for freshness can conflict with retaining consecutive samples useful for destination-side estimation.This trade-off is explicitly left outside the paper’s scope.
- Metrics: Peak age measures the maximum age immediately before receiving an update and offers a simpler formulation with a complete probability distribution.It is also related to the probability that age exceeds a threshold, relevant when freshness must be guaranteed.
- System model: The source-destination link is modeled using queuing theory, with status updates represented as timestamped packets transmitted through a finite-capacity system.The system considers a source observing a random process and sending packets containing the process value and generation time.
III. METRICS FOR AGE OF INFORMATION
This section presents average age and peak age as two metrics for characterizing information freshness.
- Average age is one of the two metrics used to characterize the age of information.
- Peak age is the second metric used to characterize the age of information.
- The metrics are introduced after examining sawtooth sample paths of the age process.
A. Average Age
Average age is obtained from the long-run time average of the age process, while peak age captures the maximum age immediately before an update arrives.
- Average Age: The average age is defined as the infinite-horizon time average of the age process, assuming ergodicity.
- Average Age: The time-average calculation represents the area under the age curve over the observation interval.
- Average Age: The average age can be decomposed into geometric areas Q_k associated with successive updates, plus a possible terminal partial area.
- Average Age: The effective arrival rate is the limiting rate of transmitted packets per unit time.
- Peak Age: Peak age measures the maximum age immediately before an update is received and can support threshold-based age analysis.
IV. CHARACTERIZING AVERAGE AGE AND PEAK AGE FOR SELECTED QUEUING MODELS
The analysis characterizes average and peak age by modeling queue-specific system times and interdeparture times, conditioning on whether a departure leaves the system empty.
- Analysis Framework: Average-age analysis requires expected system-time and interdeparture quantities, including E[T_k−1Y_k].
- Analysis Framework: Peak-age analysis requires the distribution of the sum of the previous system time T_k−1 and interdeparture time Y_k.
- Conditional Analysis: Conditioning on ψ makes T_k−1 and Y_k conditionally independent when the previous departure leaves the system empty.
- Conditional Analysis: Under the complementary event, a waiting packet begins service immediately, so the next interdeparture time is a service time independent of the previous system time.
- Selected Queuing Models: The paper studies average age and peak age across the M/M/1/1, M/M/1/2, and M/M/1/2* queuing models.
A. M/M/1/1 Queue
For the M/M/1/1 queue, packets enter only when the server is idle, and the resulting age behavior is analyzed through service and interdeparture times.
- Queue Model: The M/M/1/1 queue has two states: an empty system with probability p_0 and a one-packet system with probability p_1.
- Queue Model: Packets are accepted only when the server is idle, giving effective arrival rate λ_e = λ(1 − p_1).
- Average Age: In this model, transmitted packets spend only their exponentially distributed service time in the system.
- Average Age: As λ → ∞, the average age approaches 2/µ, the lower bound for single-server FCFS systems.
- Peak Age: The peak-age density is formed by convolving the exponential service-time distribution with the interdeparture-time distribution.
- Peak Age: The peak-age complementary cumulative distribution describes the probability that peak age exceeds a threshold.
- Peak Age: The average peak age is obtained by integrating the peak-age distribution or by combining the relevant expected service and interdeparture times.
B. M/M/1/2 Queue
The M/M/1/2 model represents the system with three Markov states and derives packet waiting, service, and system-time distributions to calculate average age.
- The M/M/1/2 queue has three states: empty, one packet in service, or one packet in service with one waiting.
- Packets are accepted whenever the system is not full, giving effective arrival rate λe = λ(1 − p2).
- A transmitted packet’s system time is decomposed as Tk−1 = Wk−1 + Sk−1, separating waiting and service time.
- PASTA makes the probability of finding the system idle upon arrival equal to the steady-state idle probability, while busy arrivals wait for residual service.Residual service is exponential with parameter µ because service times are memoryless.
- The analysis conditions system-time distributions on whether the preceding system state leaves no packet or another packet waiting after service.The no-arrival and at-least-one-arrival cases produce separate conditional expectations used in the age calculation.
2) Average Age:
The paper combines system-time and interdeparture components to obtain the M/M/1/2 average age and examines its behavior as the arrival rate grows.
- The M/M/1/2 average age is calculated from the preceding packet’s system time and the subsequent interdeparture-related quantity.
- As the arrival rate tends to infinity, the M/M/1/2 average age approaches the expression derived in the limiting analysis.
- Keeping a packet in the buffer is not advantageous at very high arrival rates because information ages while waiting in queue.The paper states that waiting for a new packet is preferable to retaining one in the buffer in this regime.
3) Peak Age:
Peak age is analyzed by conditioning on whether the system is left idle or with a packet waiting, then combining the corresponding conditional distributions.
- The peak-age distribution is a mixture formed from the conditional distributions associated with ψ and its complement.Conditionally independent Tk−1 and Yk allow the distribution of their sum to be obtained by convolution.
- The complementary cumulative distribution function gives the probability that peak age exceeds a threshold.
- The paper provides an explicit expression for the peak-age complementary cumulative distribution function.
- The expected peak age is calculated directly from the derived peak-age distribution.
C. M/M/1/2* Queue
The M/M/1/2* queue replaces a waiting packet whenever a new packet arrives, preserving the three-state occupancy model while requiring separate analysis of transmitted and dropped packets.
- In M/M/1/2*, a newly arriving packet replaces the packet waiting in queue, but the number of packets in the system remains identical to M/M/1/2.The model is therefore described by the same three-state Markov chain.
- Little’s result does not apply directly to buffer packets because some leave after waiting while others are discarded.The analysis instead characterizes the time in system for any packet and mixes idle, transmitted, and dropped cases.
- For packets arriving to a busy server, transmission occurs only if no other arrival interrupts the remaining service.
- The time in system for a transmitted packet is T = W + S, with waiting and service components analyzed conditionally on the arrival state.
- Packets dropped while waiting spend a time X equal to the interval until the next arrival, conditioned on that arrival occurring before service completion.
2) Average Age:
For the M/M/1/2* model, the average age is expressed analytically and its infinite-arrival-rate limit matches the no-buffer M/M/1/1 model. Packet replacement is asymptotically optimal and performs better at finite arrival rates for fixed service rate.
- The M/M/1/2* average age is written using equations (29) and (65).
- As λ approaches infinity, M/M/1/2* behaves like the no-buffer M/M/1/1 model.At very high arrival rates, both models effectively always have a fresh packet available for transmission.
- Packet replacement is asymptotically optimal and provides better average-age performance than M/M/1/1 at finite arrival rates with fixed service rate.
- The peak-age distribution is obtained as a mixture of conditional convolutions for the relevant events.Conditional independence of T_k−1 and Y_k enables the convolution construction.
3) Peak Age:
The paper characterizes peak age using the complementary complementary cumulative distribution and an integration-based average for queuing models with packet management.
- The conditional complementary cumulative distribution describes the probability that peak age surpasses a specified threshold.
- For the M/M/1/2* model, the paper presents an explicit expression associated with the peak-age distribution.
- Average peak age can be obtained by integrating the complementary cumulative distribution or summing the expected values of T_k−1 and Y_k.
- Figure 3 plots average age against arrival rate λ for queuing models with packet management at service rate µ = 1.
V. NUMERICAL RESULTS
Numerical results show that packet management, especially replacement, improves average and peak age relative to retaining stale packets, with benefits depending on arrival and service rates. The peak-age results further identify replacement as best across tested regimes and simple discarding as preferable at very high arrival rates.
- Experimental setup: The study evaluates analytical results with numerical examples and simulation points, which corroborate the analytical results.Unless otherwise stated, the experiments use unitary service rate µ = 1.
- Average age with packet management: Packet replacement yields the smallest average age; at λ = 0.6, it reduces average age by approximately 5% versus the other packet-management models.Average age decreases with channel utilization over 0 < ρ ≤ 1.5.
- Comparison with no packet management: Packet management eliminates the divergence of average age near λ = µ seen in the M/M/1 model, with significant improvement particularly for ρ > 0.5.At low arrival rates, both models behave similarly because they are idle for a large fraction of time.
- Service-rate comparison: With fixed λ = 0.5, M/M/1/2* has the smallest average age, while all schemes approach the common limit 1/λ as µ increases.For very small service times, average age is limited by interarrival times.
- Peak-age distribution: For peak-age distributions at µ = 1, larger arrival rates reduce the probability that peak age exceeds threshold a, and M/M/1/2* performs best for λ = 0.5 and λ = 1.3.M/M/1/2 is second best at small arrival rates, whereas M/M/1/1 outperforms it at large arrival rates.
- Average peak age: Average peak-age results favor buffering at small arrival rates, but support discarding busy-arrival packets when λ > µ; packet replacement performs best overall.The replacement model is identified as suitable when receiver age must remain below a threshold.