Harsh RED: Improving RED for Limited Aggregate Traffic

Abstract

Abstract-A bottleneck router typically resides close to the edge of the network where the aggregate traffic is often limited to a single or a few users only. With such limited aggregate traffic Random Early Detection (RED) on the bottleneck router is not able to properly respond to TCP slow start that causes rapid increase in load of the bottleneck. This results in falling back to tail-drop behavior or, at worst, triggers the RED maximum threshold cutter that drops all packets causing undesired break for all traffic that is passing through the bottleneck. We explain how TCP slow start, ACK clock and RED algorithm interact in such a situation, and propose Harsh RED (HRED) to properly address slow start in time. We perform a simulation study to compare HRED behavior to that of FIFO and RED with recommended parameters. We show that HRED avoids tail-drop and the maximum threshold cutter, has smaller queues, and provides less bursty drop distribution.

Evaluating CoDel, PIE, and HRED AQM techniques with load transients

“…Adaptive RED (ARED) [12], [15] is also likely to be too slow to respond resulting in similar behavior as with regular RED because state is only updated using relatively infrequent timer while the slow start keeps increasing the load constantly. Usually such a slow AQM response leads to fallback using tail-drop [26]. Instead of auto-tuning the dropping probability slowly such as with ARED, HRED immediately starts with aggressive dropping.…”

“…HRED [26] is an aggressive RED variant that is designed TCP slow start in mind. The recommended RED parametrization [13] is not useful with load transients.…”

“…For comparison we use HRED (Harsh RED) [26] which we built earlier on top of RED. Our study significantly differs from earlier studies on PIE and CoDel performance [28], [46], [47], [17], [18], as we focus on very short-term transient effects that are typical with the traffic patterns of today and do not introduce long-running TCP flows at all.…”

Improved QoS for Multimedia Transmission using Buffer Management in Wireless Sensor Network

“…The difficulties associated with this approach are that the overload of data packets reveals in discontinuation and jitter. Jarvinen et al [20] addressed RED (Random Early Deduction) technique in which the authors proved that it has not more benefits as it is too moderate to tackle rapid change because of slow start TCP, particularly while the data traffic is inadequate. Each and every one of above methods changes the buffer size either by incrementing or decrementing the window size.…”

Report from the IAB/IRTF Workshop on Congestion Control for Interactive Real-Time Communication

“…Random Early Detection (RED) [9] is one such AQM algorithm, but it has not been widely deployed in routers largely because of challenges to configure it correctly [32]. According to [23], RED does not work with the default settings as it is "too "gentle" to handle fast changes due to TCP slow start, when the aggregate traffic is limited." There may also be a lack of incentives to deploy AQM algorithms.…”