Most of scientists except computer scientists do not want to make efforts for performance tuning with rewriting their MPI applications. In addition, the number of processing elements which can be used by them is increasing year by year. On large-scale parallel systems, the number of accumulated messages on a message buffer tends to increase in some of their applications. Since searching message queue in MPI is time-consuming, system side scalable acceleration is needed for those systems. In this paper, a support function named LHS (Limited-length Head Separation) is proposed. Its performance in searching message buffer and hardware cost are evaluated. LHS accelerates searching message buffer by means of switching location to store limited-length heads of messages. It uses the effects such as increasing hit rate of cache on host with partial off-loading to hardware. Searching speed of message buffer when the order of message reception is different from the receiver's expectation is accelerated 14.3 times with LHS on FPGA-based network interface card (NIC) named DIMMnet-2. This absolute performance is 38.5 times higher than that of IBM BlueGene/P although the frequency is 8.5times slower than BlueGene/P. LHS has higher scalability than ALPU in the performance per frequency. Since these results are obtained with partially on loaded linear searching on old Pentium®4, performance gap will increase using state of art CPU. Therefore, LHS is more suitable for larger parallel systems. The discussions for adopting proposed method to state of art processors and systems are also presented.

An active queue management (AQM) scheme is proposed to reduce throughput bias for UDP flows over TCP. It is argued that existing AQM methods partially involve a flow-indifferent factor that does not take into account of bandwidth usage of flows when they determine packet drop, thus resulting in unfairness. The proposed scheme replaces the flow-indifferent part with a flow-wise one by approximating per-flow fair share, which permits the discrimination of unresponsive flows. Since it is a stateless process, it avoids the overhead of tracking the statistics of flows and implementation is simple. A performance evaluation shows that it effectively limits the bandwidth of unresponsive flows to their fair share of bandwidth. In addition, it can also encourage RTT-fairness among TCP flows with different delays.

Multimedia traffic on the Internet is rapidly increasing with the advent of broadband networks. However, the Best-Effort (BE) service used with Internet Protocol (IP) networking was never intended to guarantee Quality of Service (QoS) for each user. Therefore, the realization of QoS guarantees has become a very important issue. Previously, we have proposed a queue management scheme, called Dual Metrics Fair Queuing (DMFQ), to improve fairness and to guarantee QoS. DMFQ improves fairness and throughput by considering the amount of instantaneous and historical network resources consumed per flow. In addition, DMFQ has characteristics of high speed and high scalability because it is hardware oriented. However, DMFQ may be unable to adapt to network fluctuations, given that it has static setup parameters. Moreover, DMFQ is unable to support a multiclass environment. In this paper, we propose a new buffer management scheme based on DMFQ that can adapt flexibly to network conditions and can provide classified services. The proposed scheme stabilizes buffer utilization within a fixed range by controlling the buffer threshold, which affects the calculated packet discard probability. Moreover, by applying the proposed scheme to Differentiated Services (DiffServ), we achieve prioritized buffer management.

We study the interaction of TCP and the proportional fair scheduling algorithm in wireless networks. We show that the additive increase and multiplicative decrease algorithm of TCP can favor bad channel users, which results in inefficient use of radio resources. To remedy the problem, a proportional queue management scheme is proposed. The effectiveness of the algorithm is shown by simulations.

Service differentiation is one of the key issues in the current Internet. In this paper, we focus on a recent proposal for proportional loss rate differentiation which employs a single FIFO queue, an AQM algorithm for computing the packet drop probability, and a counter-based packet dropping routine for achieving the intended proportional loss rate differentiation among classes. It is first shown that, when the target dropping probability of a class is large, the counter-based packet dropping routine may yield a significant amount of error between the target and measured drop probabilities for the class, and subsequently, fails to maintain the loss rate ratios between classes as intended. To avoid this problem, a new compensatory packet dropping routine is developed in this paper. Then, a series of simulation experiments are conducted using the ns-2 simulator to assess the performances of the two dropping routines under various congestion conditions and quality spacings between classes. The simulation results show that, unlike the counter-based dropping routine, the proposed compensatory dropping routine is effective in keeping the loss rate ratios between classes closely on target regardless of the degree of congestion and quality spacing between classes, while the two dropping routines perform similarly in terms of throughput and queueing delay in the bottleneck link. In addition, such robustness of the proposed routine is achieved without any additional control parameter or computational effort compared to the counter-based routine.

In this paper we propose a cross-layer retransmission control for TCP communication over a wireless link. With our proposed control, a retransmission delay for lost packet is reduced, packet losses in the wireless link are eliminated and all packets are delivered in the correct order. No change is required to TCP itself or to the sender. Our proposed method is implemented in a queue between the media access control (MAC) layer and logical link layer in a base station, and is designed to assist local retransmission control in the MAC layer. Computer simulations show that our proposed method can maximally use the bandwidth of the wireless link under high bit error rates conditions with conventional TCP control. The fairness problem of TCP communication between connections with different bit error rates in a wireless link is also improved, and MAC level fairness is also controllable.

Several active queue management schemes have been proposed to provide fairness among flows. In particular, BLACK has the best performance among them with respect to simplicity and fairness. However, it can provoke the UDP throughput to fluctuate and worsen the fairness. In this paper, we propose a new active queue management scheme to modify BLACK for better performance. Simulation results illustrate that the performance of our proposal is better than that of BLACK in terms of the fairness and the smooth UDP throughput.

Active Queue Management (AQM) based on nonlinear difference equations is proposed to solve the end-to-end TCP network congestion problem. The proposed AQM scheme can guarantee the stability of the multiple bottleneck network by nonlinear control of dropping probability of the routers by imposing some restrictions on the AQM parameter. Nonlinear control often relies on some heuristics and network traffic controllers that appear to be highly correlated with the multiple bottleneck network status. Based on the proposed nonlinear difference equations for TCP flows control across the network, this paper reveals the reasons of congestion of multiple bottleneck AQM, and provides a theorem for avoiding network congestion. Moreover, we give simulations to verify the results for nonlinear control of the multiple bottleneck network congestion.

In this paper, based on Equivalent Active Flow, we propose a novel technique called Approximate Fairness Dropping, which is able to approximate fairness by containing misbehaving flows' access queue opportunity with low time/space complexity. Unlike most of the existing Active Queue Management schemes (e.g., RED, BLUE, CHOKE), Approximate Fairness Dropping does not drop the packets whose arriving rate is within the maximum admitted rate, so it protects the well-behaving flows against misbehaving ones, moreover, improves the throughput and decreases the queuing delay. Our simulations and analyses demonstrate that this new technique outperforms the existing schemes and closely approximates the "ideal" case, where full state information is needed.

Random Early Detection (RED), an active queue management scheme, has been recommended by the Internet Engineering Task Force (IETF) for the next generation routers. RED suffers from a number of performance problems, such as low throughput, large delay/jitter, and induces instability in networks. Many of the previous attempts to improve the performance of RED have been based on optimizing the values of the RED parameters. However, results have shown that such optimizations resulted in limited improvement in the performance. In this paper, we propose Double Slope RED (DSRED), a new active queue management scheme to improve the performance of RED. The proposed scheme is based on dynamically changing the slope of the packet drop probability curve as a function of the level of congestion in the buffer. Results show that our proposed scheme results in better performance than original RED.

This paper introduces a State Dependent Random Early Detection (SDRED) scheme that can improve delay and jitter performances by adjusting RED parameters such as maxth and wq according to queue status. The SDRED is designed to adapt to traffic conditions by adjusting the maximum threshold and queue weight to four different levels. From the simulation results, we show that the SDRED decreases queuing delay and has more stable jitter characteristics than the existing RED, BLUE, ARED and DSRED schemes.

Congestion control in the Internet consists of two main components: the TCP Additive-Increase Multiplicative-Decrease (AIMD) mechanism on sending windows implemented by end-users, and the Active Queue Management (AQM) scheme implemented in the routers which improves the effectiveness of congestion control. TCP connection is regarded as a feedback control system. Comparably, AQM is classified as a flow controller. There are several kinds of time delays in the network, such as propagation delay, queuing delay in the buffer of the router, etc. The time delays cause degradation of performance and instability of the network. A Smith Predictor is commonly used in feedback control of plants with significant time delays to implement effective compensation. In this paper, a Smith Predictor-based PI-controller for AQM (SPPA) is proposed, which uses a TCP reference model and an average Round-Trip Time (RTT) to reduce unfavorable effects of time delays in TCP networks. The drop probability is calculated by a Proportional-Integral (PI) controller based on the prediction error. When a mismatch exists in between the actual model of the TCP process and the reference model employed by the SPPA, we demonstrate conditions under which the network is stable. The performance, robustness and effectiveness of the proposed SPPA are all evaluated using simulations. The performance of the SPPA is compared with some typical AQMs, such as the Adaptive RED, the PI-controller, and the Proportional-Differential (PD) controller.

Adaptive Virtual Queue (AVQ) introduces a novel implementation algorithm for Active Queue Management (AQM). The stability criterion for AVQ was deduced in literature [1], but it lacks practicability due to the difficulty of solving the transcendental equation. In this letter, the AVQ stability is further investigated based on the characteristic roots of delay-differential equation. Another stability criterion explicitly associated with parameters of network configuration is deduced and the upper bound of delay time for stable AVQ algorithm is determined. Finally, the conclusion is validated through simulation experiments.

Recently, various types of traffic have increased on the Internet with the development of broadband networks. However, it is difficult to guarantee QoS for each traffic type in current network environments. Moreover, it has been reported that bandwidth can be allocated to flows unfairly, and this can be an important issue for QoS guarantees. Therefore, we have proposed a flow-based queue management scheme, called Dual Metrics Fair Queueing (DMFQ), to improve the fairness and QoS per flow. DMFQ discards arrival packets by considering not only the arrival rate per flow but also the flow succession time. In addition, we have confirmed the effectiveness of DMFQ through several computer simulations. In this paper, we implement DMFQ with hardware for high-speed operation. Concretely, we propose the design policies and show the hardware design results.

Active Queue Management (AQM) can maintain smaller queuing delay and higher throughput by purposefully dropping packets at the intermediate nodes. Most of the existing AQM schemes follow the probability dropping mechanism originated from Random Early Detection (RED). In this paper, we develop a novel packet dropping mechanism for AQM through designing a two-category classifier based on the Fisher Linear Discriminate approach. The simulation results show that the new scheme outperforms other well-known AQM schemes, such as RED, AdaptiveRED, AVQ, PI, REM etc., in the integrated performance. Additionally, our mechanism is simple since it requires few CPU cycles, which makes it suitable for the high-speed routers.

The window flow control based end-to-end TCP congestion control may cause unfair resource allocation among multiple TCP connections with different RTTs (round trip times) at a bottleneck link. In this paper, in order to improve this unfairness, we propose the active ECN which is an ECN based active queue mechanism (AQM). A bottleneck router with the proposed mechanism marks TCP segments with a probability which depends on the RTT of each connection. By enabling the TCP senders to reduce their transmission rate when their packets are marked, the proposed mechanism can realize the same transmission rate among TCP connections with different RTTs. Furthermore, the active ECN can directly mark ACKs from TCP receivers, while the conventional ECN marks TCP segments coming from the TCP senders. As a result, the queue length distribution at the bottleneck link gets stabilized, because the sender can quickly react to the marking according to variation of the queue length.

In this letter, we evaluate the performance of several adaptive and non-adaptive AQM schemes for congestion control in a dynamic network environment with variable bandwidth links. The AQM schemes examined are RED, BLUE, Adaptive RED, REM, AVQ and PI controller. We compare their queueing performance and show that none of them can derive stable queue length and low packet drop rate simultaneously in networks where both input traffic and available output bandwidth are time varying. Adaptive and efficient algorithms should be designed and applied in order to improve the adaptiveness and robustness of congestion control in dynamic networks such as Internet.

In this paper, we propose a nonlinear control model to characterize the AQM algorithm-GREEN. Based on this model, we analyze its performance and prove that there exists a stable oscillation when in equilibrium. Furthermore, we also investigate the effects of the factors such as bandwidth, round trip time, and load level on the amplitude and frequency of the oscillation. Theoretical analysis and simulation results indicate that GREEN algorithm is insensitive to the network conditions when the link rate and the round trip time are relatively small and becomes more sensitive to the change of network conditions when the bandwidth delay product is relatively high. For GREEN the adaptability to a wide range of network conditions is based on the compromising of the efficiency.

Evolutionary learning methods have been applied to a variety of different problems. In this paper, a new algorithm for active queue management based on an evolutionary learning model is proposed. This novel algorithm generates packet marks for the purpose of improving robustness and responsiveness of congestion control in the Internet routers, while maintaining a reasonable degree of queueing performance such as utilization, delay, and packet drops due to buffer overflow. Simulation results demonstrate the effectiveness of the proposed algorithm and compare the performance of various algorithms.

Today's market share of Intelligent Network (IN) service is growing rapidly in wireless networks due to the rapid advances in wireless telecommunication and IN technology. To guarantee network independent IN services, mobility of IN service subscribers has to be taken into account. This paper proposes new designs of Global Service Logic for the IN service enhancement, which increase call completion rates in wired and wireless intelligent networks. In order to apply this logic to wireless service subscribers as well as wired service subscribers, we implement a Queue Manager applied to the call queuing service feature in the Service Control Point (SCP). In the case of wireless service subscribers, the Home Location Register (HLR) handles the service registration flags to notify the Queue Manager of the corresponding service subscribers' mobility. In addition, we present a dynamic queue management mechanism, which dynamically manages the queue size based on the parallel server queuing model as the wireless subscribers roam the service groups due to their mobility characteristics. In order to determine the queue size allocated by the dynamic queue manager, we simulate the relationship between the number of the subscriber's terminals and the drop rate by considering the service increment rate. Moreover, the appropriate waiting time in the queue as required is simulated according to the above relationship. We evaluate call completion rates of the proposed mechanism in the paper by comparing to that of the existing mechanism.