Stability of Finite Population ALOHA with Variable Packets

ALOHA is one of the most basic Medium Access Control (MAC) protocols and represents a foundation for other more sophisticated distributed and asynchronous MAC protocols, e.g., CSMA. In this paper, unlike in the traditional work that focused on mean value analysis, we study the distributional properties of packet transmission delays over an ALOHA channel. We discover a new phenomenon showing that a basic finite population ALOHA model with variable size (exponential) packets is characterized by power law transmission delays, possibly even resulting in zero throughput. These results are in contrast to the classical work that shows exponential delays and positive throughput for finite population ALOHA with fixed packets. Furthermore, we characterize a new stability condition that is entirely derived from the tail behavior of the packet and backoff distributions that may not be determined by mean values. The power law effects and the possible instability might be diminished, or perhaps eliminated, by reducing the variability of packets. However, we show that even a slotted (synchronized) ALOHA with packets of constant size can exhibit power law delays when the number of active users is random. From an engineering perspective, our results imply that the variability of packet sizes and number of active users need to be taken into consideration when designing robust MAC protocols, especially for ad-hoc/sensor networks where other factors, such as link failures and mobility, might further compound the problem.

💡 Research Summary

**
This paper revisits the classic ALOHA medium‑access control protocol, but instead of the usual mean‑throughput analysis it focuses on the full distribution of packet transmission delays. The authors consider a finite‑population ALOHA system with M ≥ 2 users, where each user generates new packets after an exponential inter‑arrival time with mean 1/λ. Crucially, packet sizes are not fixed; they are independent random variables L with an exponential‑type tail, i.e.,
\