Tight Analysis of Priority Queuing Policy for Egress Traffic

Recently, the problems of evaluating performances of switches and routers have been formulated as online problems, and a great amount of results have been presented. In this paper, we focus on managing outgoing packets (called {\em egress traffic}) on switches that support Quality of Service (QoS), and analyze the performance of one of the most fundamental scheduling policies {\em Priority Queuing} ($PQ$) using competitive analysis. We formulate the problem of managing egress queues as follows: An output interface is equipped with $m$ queues, each of which has a buffer of size $B$. The size of a packet is unit, and each buffer can store up to $B$ packets simultaneously. Each packet is associated with one of $m$ priority values $\alpha_{j}$ ($1 \leq j \leq m$), where $\alpha_{1} \leq \alpha_{2} \leq \cdots \leq \alpha_{m}$, $\alpha_{1} = 1$, and $\alpha_{m} = \alpha$ and the task of an online algorithm is to select one of $m$ queues at each scheduling step. The purpose of this problem is to maximize the sum of the values of the scheduled packets. For any $B$ and any $m$, we show that the competitive ratio of $PQ$ is exactly $2 - \min_{x \in [1, m-1] } { \frac{ \alpha_{x+1} }{ \sum_{j = 1}^{x+1} \alpha_{j} } }$. That is, we conduct a complete analysis of the performance of $PQ$ using worst case analysis. Moreover, we show that no deterministic online algorithm can have a competitive ratio smaller than $1 + \frac{ \alpha^3 + \alpha^2 + \alpha }{ \alpha^4 + 4 \alpha^3 + 3 \alpha^2 + 4 \alpha + 1 }$.

💡 Research Summary

The paper addresses the management of egress traffic on QoS‑enabled switches by modeling it as an online scheduling problem. An output interface contains m queues, each equipped with a buffer of size B. Packets are unit‑size and each carries a priority value αj (1 ≤ j ≤ m) with α1 = 1 ≤ α2 ≤ … ≤ αm = α. At every scheduling step an algorithm must pick one queue and transmit the head‑of‑line packet, gaining a reward equal to the packet’s priority. The objective is to maximize the total reward over an arbitrary arrival sequence that is revealed online.

The authors focus on the classic Priority Queuing (PQ) discipline, which always serves the non‑empty queue with the highest priority value. Although PQ is widely deployed because of its simplicity, its worst‑case performance had only been loosely bounded (a 2‑approximation) in prior work. This paper delivers a tight competitive‑ratio analysis for PQ that holds for any buffer size B and any number of queues m.

Exact competitive ratio of PQ
Through a careful construction of worst‑case inputs, the authors show that the competitive ratio of PQ is
\