The Research Of Active Queue Management Based On Average Queue And Load

The rapid development of Internet has caused serious congestion in the network. Queue management is an effective method to implement congestion control at the network layer. Drop Tail is a queue management mechanism most commonly implemented in current network, it sets a maximum length for each queue, accepts packets for the queue until the maximum length is reached, and drops the subsequent incoming packets until the queue decreases. It is easy to implement, but it often causes problems of Full Queues and Lock-Out, thus depresses the performance of the network. Active Queue Management (AQM) notifies the source of incoming congestion by dropping packets before buffers overflow, enables the end nodes to respond to congestion before a queue becomes full, it reduces numbers of packets dropped in the routers, provides lower-delay interactive service and avoids lock-out behavior.In RFC 2309, the Internet Research Task Force (IRTF) recommends the deployment of Active Queue Management mechanisms such as Random Early Detection (RED). But the performance RED algorithm depends on the tuning of its operating parameters and it is unstable when we test it by different traffic. In this paper, we present an adaptive RED based on both the drop rate in the case of a stable traffic and the change of load on the router to improve its stability under different traffic circumstance.Most AQM mechanisms drop packet randomly or drop the incoming packet when they decide to drop a packet, it is unfair to those flows that have few packets to send or flows with long Round-Trip-Time (RTT). In this paper, we propose Selective Dropping Strategy (SDS) that estimates the RTT of each flow and uses a virtual queue to record the per-active-flow state information such as the quantity of newly incoming packets and the newly dropping time and its estimated RTT. We use a method called pseudo-round-robin to filter out flows occupied more bandwidth than mean bandwidth from those active flows registered in the virtual queue. The candidate flows are consisted of flows that occupied more bandwidth than mean bandwidth and didn't have packets be dropped by the router recently. When it is time to drop a packet, we'll drop the candidate flow's packet in the front of the queue. Therefore, the flow's packet loss is the direct ratio of the bandwidth it has occupied and the inverse ratio of its RTT.Because the average queue length is not sensitive to the fluctuation of load while using the router's load as the only measurement of congestion has the following shortcoming:careful design is required to provide good service. We propose a new AQM mechanism called Early Selective Drop (ESD), which uses both the average queue length and the router's load fluctuation as the measurement of congestion. We need give an objective queue delay for ESD only, and it use an exponential function to compute the dropping probability. Based on SDS, ESD isolates the unresponsive flows from responsive flows and punishes aggressive flows to protect short flows and long RTT flows during the congestion.Quick-Start is a modified and compatible version of TCP's Slow-Start mechanism, the purpose of this mechanism is to improve the link utilization of over-provisioned network or high-bandwidth network and reduce the response of applications. After introduce this mechanism, we mainly discuss the queue management on the router with Quick-Start, and extend the "three-way handshake" to support the queue management on the router.We use NS2 as the simulation platform to verify the research mentioned above, the simulation results show that the utilization of load in RED can improve the stability of RED and reduce its drop rate; The adoption of SDS can decrease the aggregate drop rate while enhance the stability and shorten the application's response time; ESD can provide better stabilities and fairness than RED-PD, RED and Drop Tail.