Adaptive RED for FreeBSD: Design, Implementation and Challenges

Abstract

Bufferbloat problem arises due to buffering of large amounts of data in queues, owing to the large size of these queues. Bufferbloat being a relatively new phenomenon meant that earlier queue management algorithms did not specifically address this problem. Despite this issue, there is merit in analysing and evaluating old queue management algorithms which have helped alleviate the undesirable performance issues that arose due to persistently full buffers. One of the earliest and most significantly studied Active Queue Management (AQM) algorithms is Random Early Drop (RED). RED helps to keep the average size of the queues low and allow occasional bursts of packets through the queue. Once the number of packets queued crosses a minimum threshold, incoming packets are dropped with a random probability. However, the resulting average queue length is quite sensitive to the level of congestion and the RED parameter settings. Adaptive RED (ARED) solves most of the issues faced by RED with minimal changes and leaves its basic idea intact. The ARED algorithm regularly adapts the value of the maximum dropping probability and ensures that the queue length stays within the targeted range. Despite its ability to resolve the inherent problems in RED, ARED went largely unnoticed for several years, until the issue of Bufferbloat arose. Although ARED predates Bufferbloat, its fundamental design makes it an effective solution to handle Bufferbloat. This discovery led to the implementation of ARED in Linux and in network simulators like ns-3. Besides Linux, FreeBSD is one of the most popular open source operating systems. Although RED is supported in FreeBSD, ARED is not. Since ARED is one of the viable solutions to tackle Bufferbloat, this paper discusses the design and implementation of ARED in FreeBSD. We also detail the challenges faced during the implementation, and validate through real testbed experiments that our implementation in FreeBSD exhibits ARED's key characteristics. � 2019 IEEE.