This letter proposes a combined input- and crosspoint-queued (CIC) switch in which virtual output queuing (VOQ) is used at each input port. This CIC switch has a large buffer at each input port and a small buffer at each crosspoint. It does not require high-speed memory access or high-speed internal cell transmission lines. Since the performance of the CIC switch depends on the scheduling algorithms, we propose new scheduling algorithms for the CIC switch. Numerical results show that the mean cell delay time performance of the CIC switch using the proposed scheduling algorithms is better than that of an input-queued ATM switch. In addition, the required buffer size for the CIC switch using the proposed scheduling algorithms is smaller than that for a crosspoint-queued ATM switch.

Even though information in ATM networks is handled as fixed-sized packets (cells), packet-based scheduling is still needed in ATM networks. This letter proposes a packet-based scheduling mechanism that is based on comparison between a packet-based queue and a virtual queue that represents the queue length provided by a cell-based scheduling mechanism. Simulation results showed that this proposed scheduling allocates the bandwidth fairly to each connection.

It is reported that TCP does not perform well in high-speed wide area networks. Because MulTCP behaves like the aggregate of N TCP flows, MulTCP can be used to achieve throughputs of 1 Gbps or more. However, no performance evaluation of MulTCP in high-speed wide area networks has been published. Computer simulations are used to evaluate the performance of MulTCP. The results clarify that synchronized packet losses greatly impact the performance of MulTCP.

A multi-QoS scalable-distributed-arbitration (MSDA) ATM switch is described that supports both high- and low-priority traffic under the head-of-line-priority discipline. It uses crosspoint and transit buffers, each consisting of a high- and low-priority buffer. The buffers arbitrate in a distributed manner the selection of which cellsto transmit. The MSDA switch supports multiple QoS classes while still providing the scalability of a previously described single-QoS scalable-distributed-arbitration (SSDA) switch. A problem occurs when the delay-time-based cell-selection mechanism used in the SSDA switch is applied to the low-priority traffic: it cannot achieve fairness in terms of throughput. This problem is overcome by introducing a distributed-ring-arbiter-based cell-selection mechanism at each crosspoint for the low-priority traffic. The low-priority transit buffer at each crosspoint has virtual queues, one for each upper input port. Cells for the low-priority traffic are selected by distributed-ring arbitration among the low-priority crosspoint buffer and these virtual queues. For the high-priority traffic, the same delay-time-based cell-selection mechanism is used as in the SSDA switch. Simulations show that the MSDA switch ensures fairness interms of delay time for the high-priority traffic and ensures fairness in terms of throughput for the low-priority traffic.

This paper proposes the iterative quasi-oldest-cell-first (i-QOCF) scheduling algorithm, a new scheduling algorithm for input-queued ATM switches with virtual output queuing (VOQ). In the i-QOCF scheduling algorithm, each input port and each output port maintains its own list. The length of the list can be N, 2 N, ..., B N, where B is the size of the separate queue for an output port at input ports, and N is the number of output ports. The list maintained by an input port contains the identifiers for those output ports to which that input port will send a cell. The list maintained by an output port contains the identifiers for input ports that have a cell destined for that output port. If we use a list whose length is B N, then the identifiers in the list appear in the oldest order, and i-QOCF gives preference to cells that have been waiting for the longest time. If we use a list whose length is less than B N, then the identifiers in the list appear in the quasi-oldest order, and i-QOCF gives preference to cells that have been waiting for the quasi-longest time. We determine the performance of i-QOCF in a comparison with i-OCF in terms of cell delay time. We find that an input-queued ATM switch with i-QOCF and VOQ can achieve 100% throughput for independent arrival processes. Under uniform traffic, 3-QOCF is enough to achieve convergence during one cell time. If we use 3-QOCF, the list length is 3 N, then its cell delay time is almost the same as that of 4-OCF (Oldest-Cell-First).

The ATM Forum specifies several fairness criteria, thus the scheduling mechanisms should allocate enough bandwidth to each connection to achieve one of such fairness criteria. However, two fairness criteria (MCR plus equal share, maximum of MCR or Max-Min share) cannot be achieved by conventional scheduling mechanisms. In this letter, we have developed new scheduling mechanisms that achieve these fairness criteria. We also present simulation results to show that our mechanisms can allocate bandwidth fairly.

It is well known that TCP does not fully utilize the available bandwidth in fast long-distance networks. To solve this scalability problem, several high speed transport protocols have been proposed. They include HighSpeed TCP (HS-TCP), Scalable TCP (S-TCP), Binary increase control TCP (BIC-TCP), and H-TCP. These protocols increase (decrease) their window size more aggressively (slowly) compared to standard TCP (STD-TCP). This paper aims at evaluating and comparing these high speed transport protocols through computer simulations. We select six metrics that are important for high speed protocols; scalability, buffer requirement, TCP friendliness, TCP compatibility, RTT fairness, and responsiveness. Simulation scenarios are carefully designed to investigate the performance of these protocols in terms of the metrics. Results clarify that each high speed protocol successfully solves the problem of STD-TCP. In terms of the buffer requirement, S-TCP and BIC-TCP have better performance. For TCP friendliness and compatibility, HS-TCP and H-TCP offer better performance. For RTT fairness, BIC-TCP and H-TCP are superior. For responsiveness, HS-TCP and H-TCP are preferred. However, H-TCP achieves a high degree of fairness at the expense of the link utilization. Thus, we understand that all the proposed high speed transport protocols have their own shortcomings. Thus, much more research is needed on high speed transport protocols.

It has been shown that virtual output queuing (VOQ) and a sophisticated scheduling algorithm enable an input-queued switch to achieve 100% throughput for independent arrival process. Several of the scheduling algorithms that have been proposed can be classified as either iterative scheduling algorithms or symmetric crossbar arbitration algorithms. i-OCF (oldest-cell-first) and TSA (two step arbiter) are well-known examples of iterative scheduling algorithms and symmetric crossbar arbitration algorithms, respectively. However, there are drawbacks in using these algorithms. i-OCF takes long time to find completely a conflict-free match between input ports and output ports because it requires multiple iterations. If i-OCF cannot find a conflict-free match completely, the switch throughput falls. TSA has the possibility that it finds a conflict-free match faster than i-OCF because it does not need any iterations. However, TSA suffers from the starvation problem. In this paper, we propose a new scheduling algorithm. It uses two schedulers, which we call scheduler 1 and scheduler 2, in parallel. After cells were transmitted, the information that input port i granted the offer from output port j in scheduler 2 is mapped to scheduler 1 if and only if input port i has at least one cell destined for output port j. If the information is moved, input port i and output port j are matched in scheduler 1 at the beginning of the next time slot. Our proposed algorithm uses one scheduler based on TSA and the other scheduler based on i-OCF. Numerical results show that the proposed scheduling algorithm does not require multiple iterations to find a conflict-free match completely and suffer from the starvation problem for both uniform and bursty traffic.

The problems caused by misbehaving flows are becoming important issues in high-speed best effort networks. In this paper, we propose the MXQ (MaXimal Queuing) mechanism which correctly identifies and adequately penalizes misbehaving flows. Identification and penalization are the keys to controlling misbehaving flows, which is believed to be crucial for providing best effort services to a large number of residential customers at reasonable cost. The proposed mechanism consists of estimating incoming traffic at network edges and selective packet discard at network nodes. This combination realizes the identification and penalization in a correct, adequate and easy to understand way, thus maintaining the stability and efficiency of best effort networks. A number of experiments are performed on a prototype system to examine the unfairness caused by different TCP implementations, which is one type of misbehavior. The results show that the MXQ mechanism can adequately penalize the misbehaving flows, and can improve fairness, even when differently implemented TCP flows are present.

This paper shows new techniques to construct a service network which realizes responsive large-size data transmission for widely distributed mass users. We set our service target as transferring mega-byte scale data from a server to a client within one second. ATM is recognized as a powerful technology with which to construct a wide area network infrastructure that supports multiple bandwidth services. Our fundamental principles in developing such a service network are as follows: a) The bandwidth sharing mechanism should be of the best effort rather than resource reservation type. This is because only best effort schemes remove bandwidth reservation/release overheads. b) More than a 100 Mb/s data transmission rate should be supported throughout data transfer. c) Data transfer should be completed within the round trip through the network (or a small multiple thereof). This is necessary to minimize the effect of transmission time in large-scale networks. d) The user network interface should be simply defined to allow independent evolution of both network and terminal technologies. e) Congestion control must block the spread of congestion within the network. Based on these principles, we propose the "ATM superpacket network (ATM-SN)" as the service network to realize our target service. Key techniques are as follows. (1) Best effort and cut-through transmission of superpackets whose length reaches ten mega-bytes. (2) Network nodes with large-capacity buffer memories that prevent superpacket collisions. (3) Superpacket admission control at network nodes to prevent cell overflow. (4) Superpacket-based congestion control. Our proposal assumes the existence of a high-quality ATM infrastructure that can provide a large bandwidth with a high-quality DBR cell transmission capability (cell loss ratio is less than 10E-7) and small bit error ratios (less than 10E-10). First, we detail our proposal of the ATM-SN. Next, we propose a superpacket-based congestion control technique coupled with a simple Usage Parameter Control function. We then show the evaluation results of those key techniques to confirm the effectiveness of the superpacket network.

The differentiated services (diffserv) architecture has been proposed for implementing scalable service differentiation in the Internet. Expedited forwarding and assured forwarding have been standardized as Per-Hop Behaviors (PHB) in diffserv. Assured forwarding can be utilized to realize the service, which provides each user with a minimum guaranteed rate and a fair share of the residual bandwidth. We call it guaranteed rate (GR) service. With GR service, each packet for flow i is marked in or out based on comparison between the sending rate and the minimum guaranteed rate. When congestion occurs in networks, out packets are dropped more aggressively than in packets. Recently, several fair queuing schemes have been proposed for core stateless networks. They can achieve fairer bandwidth allocation than random early detection (RED). However, there have not been any studies that consider in/out bit usage to support GR service. This paper proposes how to extend the schemes that have been proposed for core stateless networks to allow the support of in/out bit usage. We present the performance of one of the extended schemes and compare the scheme to RED with in/out bit (RIO) in terms of fair bandwidth allocation.