AQMs

AQMs, short for Active Queue Management, are a family of techniques used in network routers and switches to manage packet queues and control congestion before buffers become full. By actively marking or dropping packets (or signaling with ECN) as congestion builds, AQMs aim to keep queuing delays low, prevent bufferbloat, and avoid synchronized bursts of retransmissions that can hurt performance.

How AQMs work varies by algorithm, but common principles include monitoring the length or occupancy of a queue and applying a probabilistic drop or mark as traffic increases. The goal is to keep the average queue length near a target level and to distribute losses or marks more evenly across flows, rather than allowing a sudden, large queue build-up. Marking packets with ECN lets endpoints slow down without incurring packet loss, when both ends support ECN.

Key algorithms in the AQM family include methods that drop or mark with a probability that grows as the queue fills, as well as more recent designs that explicitly bound sojourn time or incorporate per-flow fairness. Examples range from traditional methods that aggressively preemptively drop to more modern approaches that integrate with fair queuing to ensure fair treatment of concurrent flows. Some AQMs are designed for data centers and high-speed links, while others target general internet traffic and residential networks.

AQMs interact with transport protocols such as TCP, which respond to packet loss or ECN marks by reducing sending rates. Their performance depends on configuration and traffic characteristics; improper tuning can negate benefits. Overall, AQMs provide a proactive approach to congestion management, aiming to reduce latency, improve throughput, and curb excessive queuing delays across diverse networks.