High-performance and high-precision flow monitoring is a crucial function for network management, network bandwidth usage accounting and billing, network security, network forensics, and other important tasks. Nowadays, many commercial switches/routers provide either sFlow, NetFlow, or IPFIX scheme for monitoring the flows traversing a network. sFlow is a scheme widely supported by many switches/routers due to its using a sampling-based method, which greatly reduces the CPU processing load on a switch/router and the network bandwidth required to transmit flow data to a remote collector. However, many small flows may go undetected and the estimated flow data (e.g., the packet count and byte count) for detected flows can significantly deviate from their ground truth. NetFlow, which is Cisco Systems’ proprietary technology, does not use a sampling-based method by default. Instead, it tries to collect complete and correct flow data for every flow. However, as the link speed and the flow arrival rate continue to increase, NetFlow also provides a sampling-based option to reduce the CPU utilization of the switch/router. Because NetFlow is proprietary, an Internet Engineering Task Force (IETF) working group has defined IPFIX as an open flow information export protocol based on NetFlow Version 9. The requirements for IPFIX are defined in the RFC 3917 standards. Basically, IPFIX is the same as NetFlow Version 9. Due to its high demand on the CPU of the switch/router, currently NetFlow is supported only on very high-end switches/routers and its design and implementation on these commercial switches/routers are not published in the literature. In this paper, we design and implement a high-performance and high-precision NetFlow/IPFIX system on a Programming Protocol-independent Packet Processors (P4) hardware switch. Based on a 20 Gbps playback of a packet trace gathered on an Internet backbone link, experimental results show that our novel method significantly outperforms the typical design and implementation method of NetFlow/IPFIX on a P4 hardware switch. For example, for the number of detected flows during the trace period, our method outperforms the typical method by a factor of 5.72. As for the number of flows whose packet and byte counts are correctly counted, our method outperforms the typical method by a factor of 8.57.

The Optical Proximity Correction (OPC) technique is used extensively in manufacturing masks, which are used to make Integrated Circuits (ICs). Given the high density and complexity of today's IC designs, an OPC job can easily take many days of CPU time to finish. To reduce the turnaround time of an OPC job, we have constructed a cloud for performing OPC in parallel by multiple worker nodes. This cloud is managed by the Kubernetes cloud operating system and is composed of multiple worker nodes each equipped with a GPU. Because the disparity in the performance-to-cost ratios of different GPU models is large, we derived a cost-performance model to optimally allocate GPU resources to best meet an OPC client's budget and time-to-market demands. To optimally shorten the turnaround time of an OPC job, we also designed a two-tiered method to schedule the workloads of an OPC job among these worker nodes. Experimental results show that despite the large GPU per-formance differences among worker nodes, our method enabled all worker nodes to finish their assigned workloads at almost the same time. As a result, the OPC client need not waste much time for obtaining the complete results of an OPC job. For example, when our method was used on our 13-node hybrid cloud, the difference between the longest and shortest execution time of the worker nodes divided by the longest execution time of the worker nodes was only 1.91 %.

Modern high-speed networks face critical performance challenges from microbursts—transient traffic spikes that degrade latency-sensitive applications despite low average utilization. While existing approaches detect and mitigate microbursts, they lack fine-grained control at individual flow levels. We present MicroFlow, a system that advances microburst management through per-flow monitoring and targeted priority-based migration in programmable data planes. MicroFlow dynamically adjusts scheduling priorities during congestion events without requiring packet detouring or bandwidth-consuming recirculation. Our implementation on programmable switches demonstrates significant improvement in congestion management, reducing queuing delays for short flows by 82% compared to traditional QoS mechanisms under mixed workload conditions, while maintaining full link utilization. MicroFlow enables consistent low-latency performance for critical applications during extreme traffic variations without introducing additional packet processing overhead or consuming extra bandwidth resources.

In this work, we use a Programming ProtocolIndependent Packet Processors (P4) hardware switch to design and implement a remote desktop monitoring system. This system can real-time monitor the screen of another person's remote desktop session if the session is not encrypted or if the session is encrypted but the encryption key has been compromised. Our system incurs only an average delay of 12 ms on the mouse and keyboard operations that the user performs, which can hardly be noticed. In this paper, we present the design, implementation, demonstration, and performance of our system.

A digital twin (DT) is a virtual system that represents a physical system. By simulating the behavior of a digital twin in what-if conditions, the operators of a physical system can predict the outcome of the physical system in such conditions and take advance measures to improve its performance. In this work, we built the digital twin of a large and high-speed network of an Internet Service Provider (ISP) in Taiwan and used simulations to predict how the packet loss rates of its flows would change after physical link failures, under the protection of Link Aggregation Group (LAG). To do so, we first used sFlow to collect the data of the flows traversing the network and then used the flow data to conduct the simulations. This paper presents the methodology that we developed to simulate a Network Digital Twin (NDT), its simulation results, and the insights that we obtained from this work.

A distributed coordination system (DCS) provides many useful services such as in-memory data store and distributed synchronization of data. ZooKeeper, which is being actively developed by the Apache Software Foundation as a project, is one of the most popular DCSes in the world. ZooKeeper uses multiple unicast TCP connections to forward the submitted data to all servers in its cluster, using one TCP connection for each server. However, this design will greatly waste link bandwidth when the number of servers in the cluster is increased for providing higher read performance for a large number of clients. In this work, we integrate the IP multicast capability of a hardware switch with an IP multicast software tool to replace these multiple TCP connections. Experimental results show that when 7 servers are used to build ZooKeeper, the speedup of the throughput of our approach (in terms of requests per second) against ZooKeeper can be up to 5x.

Network File System (NFS) is a distributed file system that allows clients at different locations to access files stored on a remote server as if they were stored locally. Due to its usefulness, nowadays many Network Attached Storage (NAS) devices support NFS. NFS uses the Remote Procedure Call (RPC) protocol to transmit a client's request to the server and then receive the reply from the server. Using RPC to transfer a file is like using the Stop-and-Wait protocol to transfer a file, of which the throughput will inversely decrease when the round-trip-time (RTT) between the client and server increases. In this work, we use a Programming Protocol-independent Packet Processors (P4) hardware switch to periodically prefetch "N" data blocks during a read file transfer initiated by a client. Functionally, our approach is like using the Go-Back-N protocol to transfer a file and thus outperforms the original NFS significantly. Experimental results show that when the RTT between the client and server is around 1.2 ms, which is commonly seen on a local area network, our approach achieves a throughput speedup of 3x when the file transfer size is 100 MB.

To identify human movement health problems effi-ciently and comfortably, we set up a 6-minute Timed Up and Go (TUG) test field in a hospital and used a depth camera to record the movement behavior of 60 patients during their tests. With the use of a pose landmark detection tool named MediaPipe on these recorded videos, we obtained the screen coordinates of 33 key body locations of a human body during each 6-minute TUG test. During each test, a physical therapist used her expertise to assess whether the patient had some specific symptoms. In this work, we propose a novel data processing method to more effectively use this 60-patient small data set to train machine learning models, which are later used to predict whether an individual has a specific symptom. Experimental results show that, compared to the traditional data processing method, our data processing method combined with the majority-voting prediction method significantly increases the prediction accuracy of four studied machine learning models.

In a smart city, many different kinds of smart meters (i.e., IoT devices) are used to help govern the city and their data need to be periodically collected for billing or analytics purposes. In our previous work (Wang et al., 2021), we designed and implemented a drone-based wake-up and data collection system for smart meters by using wake-up radios (WuR) and WiFi. Using this system, in this work we use the deep reinforcement learning (DRL) method to train and periodically re-train the drone agent so that it can optimally collect the data of smart meters deployed on a field. For a specific neighborhood, our method uses the results of many real-life wireless transmission measurements in the neighborhood as the initial input to the DRL method. As a result, the trained agent can direct the drone to finish the data collection task of the neighborhood more efficiently. We used the actions recommended by the trained agent to direct the drone to collect the data of ten smart meters deployed on our campus. Experimental results show that the trained agent enabled the drone to finish the data collection task with a much shorter flying distance, as compared with other methods that do not consider real-life neighborhood-specific geographic effects on wireless communication . For example, compared with a greedy method, our method without re-training the drone agent can reduce 19.7% of the flying distance. In each real-life data collection task, our method also records the actual wake-up and data collection success/failure results of all smart meters and periodically uses a batch of these data to re-train the drone agent. Experimental results show that such a periodic re-training design can continue to improve the performance of the drone agent for a specific neighborhood. For example, compared with our method without re-training the drone agent, our method with re-training the drone agent can further reduce 14.6% of the flying distance.

Multicast is an approach by which a packet sent by a sending host can be received by multiple receiving hosts. Traditional methods of performing multicast is via storing multicast routing information in the multicast group table of every switch (or router) on the multicast tree. When receiving a multicast packet, a switch/router will replicate the packet and forward the replicated packets out of the switch/router based on the information stored in its multicast group table. In RFC 5058, the authors proposed the concepts of Explicit Multicast (Xcast) in which the unicast addresses of the receiving hosts of a multicast group are carried in the header of the multicast packet. The concepts of Xcast were proposed in 2007 without design and implementation details for real hardware switches/routers. Recently, due to the emergence of programmable switches, designing and implementing Xcast in hardware switches becomes feasible. In this work, we design and implement an Xcast-like scheme in P4 hardware switches and evaluate its performance. Experimental results show that our scheme can multicast packets at the 100 Gbps line rate without losing any packet inside the switch due to pipeline processing backlogs.

Nowadays, when a user downloads a web page, many targeted or untargeted advertisements are embedded into the downloaded web page. These advertisements slow down the page download and display and occupy a very large portion of the screen. Besides, they unnecessarily consume more network bandwidth. To solve these problems, in this paper we design and implement a method inside a P4 hardware switch to block trackers and advertisements. Our method successfully blocks their operations by blocking the DNS requests issued by them. Our method is to be deployed at a switch that connects an edge network to the Internet. Using such a configuration, all devices on the edge network will be automatically protected by our method without the need to install a per-device protection software. Experimental results show that our method can block trackers and advertisements for all devices on an edge network, save the network bandwidth consumed by them, and enable a web browser to download and display web pages faster.

In the past, several time synchronization methods have been proposed for parallel and distributed simulation (PDS). Among them, one widely used conservative method is the Chandy-Misra-Bryant (CMB) algorithm. In the CMB algorithm, many null messages may be exchanged among logical processes to advance their clocks so that deadlock will not occur among them. In this work, using a data-plane programmable P4 hardware switch, we design and implement a data fusion-based approach inside the packet processing pipeline of the P4 switch. Our approach extracts the timestamp carried in exchanged null messages, computes the fusion results of these timestamps, drops unnecessary null messages inside the switch, generates new messages carrying the fusion results, and sends these generated messages to only the logical processes that can benefit from receiving these messages. Experimental results show that on an 8-host testbed, our approach can speed up a PDS by a factor of 2.75 and 1.65 when compared with the unicast and multicast approaches, respectively.

Nowadays, attacks coming from the Internet are posing serious threats to the hosts in an institution, campus, company, etc. The Network Address Translator (NAT) is a device that allows a host in a private network to interact with the hosts on the public Internet. Due to the property of NAT, unless a host that is behind a NAT actively contacts a host on the Internet, hosts on the Internet cannot actively reach the host behind the NAT. In this work, we exploit NATs and propose a multi-level hierarchical NAT scheme to protect and enhance the security of a private network. We have designed and implemented our scheme over P4 programmable hardware switches. Experimental results show that our scheme functions correctly and provides high throughput, low latency, and high stability. In addition, according to our tests, our scheme works correctly with most existing network applications.

In this paper, we design, implement, and evaluate the performance of RackNFV - a high-performance, bandwidth-efficient, and reconfiguration-agile Network Function Virtualization (NFV)/Service Function Chain (SFC) system that operates in a rack. In RackNFV, multiple servers on which Virtual Network Functions (VNFs) are executed and a hardware switch that connects these servers are co-located in a rack. The switch is dynamically partitioned into multiple slice switches that can be flexibly inter-connected to form any desired network topology. The VNF servers can be dynamically connected to the slice switches to form any desired SFC for packet flows. Packets passing through a RackNFV system can be steered through any desired SFC by traversing these slice switches and VNF servers. RackNFV uses the switching capacity of a hardware switch to achieve high-performance and bandwidth-efficient packet forwarding along a chain of VNF servers. By using a novel port-mapping technique, RackNFV can quickly conFigure a new SFC in just 400 milliseconds. Operating in a rack, RackNFV greatly saves the bandwidth in a production network and makes the deployment and maintenance of NFV/SFC faster and easier.

Recently, NB-IoT, LTE Cat-M1, Sigfox, and LoRa have been proposed as promising Low-Power Wide Area Networks (LPWANs) technologies for Internet-of-Things (IoT) applications. These technologies are aimed for IoT applications such as smart meters that can tolerate long transmission delays and only need a narrow band to periodically transmit a small amount of data. Although the performances of these technologies have been studied or compared in the literature, most comparisons were conducted in non-mobile conditions. In this work, we used drones flying at 70 km/hr along a 10 km motorway to do the performance measurements. In this paper, we report and compare the performances of these technologies in such high-speed mobile conditions in the air.

Metering the traffic of a flow and dropping the portion of traffic that exceeds the target rate set by the meter is very important to provide quality of service (QoS) in a network. Most existing switches on the market provide the meter function. Although these meters regulate the traffic of a UDP flow very well, according to our tests, they are not TCP-friendly. For example, we found that a TCP flow passing the meter of a commercial switch can only achieve about 10% of the target rate. Based on our studies, we found that this poor performance is due to the bad interactions between the TCP congestion control and the meter function. In this paper, we design and implement a TCP-friendly meter in the packet processing pipelines of a P4 switch. Experimental results show that our meter regulates a TCP flow very well and can maintain its achieved rate within 5% of the target rate. Compared with the meters in several hardware switches, our meter improves the achieved rate of a TCP flow by almost 85% of the target rate.

Tracking the rules matched by the packets of a flow when they traverse multiple switches in a network is very important and useful. In this paper, we design and implement a bandwidth- efficient In-Band Network Telemetry (INT) system that can track the rules matched by the packets of a flow in real time or in the past. We have evaluated the real performance of our system on a testbed composed of several hardware P4 switches. Our experimental results show that the system works correctly and its traffic reduction scheme can reduce the rate of generated INT reports by a factor of 39, 394, 2,055, or even up to 12,500, depending on the type of network states monitored and the threshold used to filter out less important INT reports.

Sigfox is a new LPWAN (Low-Power Wide Area Networks) wireless technology proposed for IoT applications. Its transmit power is only 22.5 dBm, but its wireless transmission range can be up to 30 km in open space. In this paper, we propose a new machine learning-based localization method to predict the location of a Sigfox module. Our method divides the space around each base station into a few sectors and trains a more accurate path-loss-to-distance model for each sector to reflect its site-specific multipath propagation environment. Our experimental results measured in a big city show that the maximum distance errors of our method are much smaller than those of the official Sigfox localization service in many cases.

In this paper, we design, implement, and evaluate the functions and performance of a parallel running and comparing system that is built on OpenVswitch software switch. Parallel running is a very important method to change from an existing system to a new system. The conversion usually occurs when the technology of the old system is outdated and a new system needs to be installed to replace the old one. During parallel running, both the old and new systems are running side by side and their behavior are compared against each other until the users are certain that the new and old systems behave exactly the same. At that time, the old system will be removed and the new system will totally replace the old system. In this paper, we use the OpenVswitch to design and implement a comparing system for parallel running. Our system has many useful features and can be applied to real mission-critical systems.

Changes in the human body weight may be caused by diseases, overeating, lack of exercise, or other health problems. In this paper, we built a prototype of a pair of weighing shoes to enable people to constantly monitor their weight changes so that they can detect and treat their health problems earlier. We placed several thin and flexible pressure sensors under the shoes and used machine learning techniques to learn the relationship between their values and the actual weight of the human body being measured. The trained model then was used to predict the weight of a human body. In our experiments, we varied the number of used pressure sensors to study its impact on the prediction error. We found that the prediction error of the trained model could be controlled within 5.8% when 8 or more pressure sensors were used for the current prototype.

LoRa/LoRaWAN is an LPWAN (Low-Power Wide Area Network) wireless technology proposed for IoT applications. It has the advantages of long-range transmission (more than 10 Km in open space) and low-power consumption, and has been used in many IoT applications such as smart metering and intelligent agriculture. The maximum payload size of a LoRaWAN packet is very short and is only 50 bytes. Thus, currently TCP/IP packets cannot be transmitted over a LoRaWAN network and a network manager cannot use common commands such as telnet or ssh to remotely log into a LoRaWAN device to troubleshoot it through its LoRaWAN network. Instead, the network manager needs to go to the remote site in person to do the troubleshooting work, which incurs a great amount of time, efforts, and costs. In this paper, we design and implement a system to transmit TCP/IP packets over a LoRaWAN network. Across the Internet, a network manager can use our system to remotely troubleshoot a LoRaWAN device via its LoRaWAN network.

In this paper, we design, implement, and evaluate the real performance of running multiple Snort IDS (intrusion detection system) VNFs (virtual network function) inside a bare metal commodity switch. In the past, normally people ran Snort on a stand-alone server and configure switches to direct packets to it for inspection. However, more recently there is a trend to implement and run VNF directly inside a switch for immediate and intelligent processing of packets. Our work of running Snort directly inside a bare metal commodity switch as a VNF is the first work of its kind in the world. In this paper, we present real performance results and important findings from this innovative work.

The LoRaWAN network is proposed for IoT applications and has the advantages of long range transmission and low power consumption. However, it suffers from very low data throughputs and long transmission duty cycle delays. In our real on-campus LoRaWAN deployment that has been lasting for one year, we found that even though we placed LoRaWAN end devices (nodes) at a short distance such as less than 700 meters from a LoRaWAN gateway, the packet loss rates at some locations are still too high to be acceptable. In this paper, we propose two approaches to overcome this high packet loss rate problem. We modified the IoT application running on nodes to let a node retransmit its lost data in multiple packets and help transmit its neighboring node's data. Our experimental results show that our approaches can effectively increase the application-layer message delivery success rate of LoRaWAN.

Recently, LoRa/LoRaWAN has been proposed as a promising LPWAN technology for Internet-of-Things (IoT) applications. For monitoring the long-term PM2.5 air quality on our campus, starting from February 1 2017, we have set up a LoRaWAN-based PM2.5 air quality monitoring system on our campus. In total, twenty-two PM2.5 sensor devices have been operating at the rooftop of buildings or on the lamp poles since then. Each of these PM2.5 sensor devices uses a LoRaWAN module to send its most recent PM2.5 value to the LoRaWAN gateway every 72 seconds. After being received successfully by the LoRaWAN gateway, the packets carrying these values were stored into a database with their SNR/RSSI information. With the sequence numbers added by our PM2.5 application running on each device, the packets stored in the database provide many important information about the long-term packet loss rate performance of LoRaWAN on campus. In this paper, we process these data to report real long-term performance of LoRaWAN when used on campus.

In this paper, we design, implement, demonstrate the functions, and evaluate the performance of MiniReal - a real SDN network testbed built over an SDN bare metal commodity switch. Using MiniReal, one can partition an SDN bare metal commodity switch into several SDN slice switches and use Ethernet cables or optical fibers to connect these slice switches together to form any desired network topology. It is very quick and easy to change the formed network topology from one topology to another topology. Once the initial wiring is done, our software can change the network topology in less than 1 second without physically rewiring any cable between slice switches.

Today, storage area networks (SAN) are widely used in data center networks to connect servers to data storage devices. In such a network, ensuring zero packet loss is very important for a server's network file system (NFS) to quickly access data across the network. Otherwise, the time spent on detecting packet losses and then retransmitting the lost data will severely harm the data access throughput and delay performance of the network file system. Recently, the SDN technology has been widely used in data center networks. In an SDN network, the SDN controller has a global view of the states of the overall network. Exploiting this capability, we designed SDN-assisted rate-based and credit-based congestion control schemes to ensure zero packet loss in the network. We used both greedy traffic and bursty NFS traffic to compare the performances of the two schemes under many different situations. In this paper, we report their relative strength and weakness and show that the SDN-assisted credit-based scheme can ensure zero packet loss while maintaining a high link utilization.

In this paper, we designed, implemented, and evaluated the performance of a Packet-in packet buffer system in an SDN bare metal commodity switch and Open vSwitch. In our approach, only the first packet of a new flow needs to be sent to the SDN controller via a Packet-in packet and all subsequent packets can be temporarily stored in the Packet-in packet buffer system until the arrival of the flow rule. Our experimental results show that our approach effectively prevents packets of a new flow from being dropped when they pass an SDN switch, significantly reduces the switch CPU usage, greatly reduces the control plane bandwidth usage, and greatly reduces the Packet-in packet processing load imposed on the SDN controller.

In this paper, we implemented OpenFlow scheduled bundle in the Open vSwitch software switch and evaluated its performance in SDN testbed networks under different scenarios. The scheduled bundle is a very new mechanism proposed in the OpenFlow protocol for achieving consistent updates among multiple SDN switches. However, this feature has not been supported in any hardware SDN switches or any software SDN switches such as the very popular Open vSwitch software switch. As a result, no results about the performance and effectiveness of scheduled bundle have been reported in the literature. In this work, we implemented scheduled bundle in Open vSwitch, which is the first implementation of scheduled bundle in the world, and evaluated its performance and effectiveness under various scenarios.

To promote smart campus, National Chiao Tung University is deploying several location-based IoT applications based on wireless communications technologies including LoRa and NB-IoT. The applications are indoor/outdoor environment conditions monitoring (such as PM2.5, temperature, and humidity), emergency buttons, and so on. This paper uses PM2.5 measurement application as an example to present the performance of LoRa when it is used on campus. We investigated the LoRa transmission performance from LoRa end-devices to the LoRa gateway. We show how packet losses were affected by the distance between the end-device and the gateway, the transmit power, the payload length, the antenna angle, the time of day, and the weather conditions. The pattern of LoRa packet losses were also measured and we found that in our long-term PM2.5 monitoring application, more than 99\% of LoRa packet losses occurred with three or less consecutive packet losses.

In this paper, we designed, implemented, and evaluated the performance of software OpenFlow flow counters in a bare metal commodity switch. Normally, flow counters are implemented in hardware for line-rate operations and their number needs to be very large to support a large flow-based SDN network. Although these hardware counters operate very fast, they greatly increase the cost of an OpenFlow switch. In addition, due to the limited chip size of the switching ASIC used in an OpenFlow switch, the number of hardware counters cannot scale to a large number. To overcome these drawbacks, we designed and implemented software flow counters in a 48 port × 10 Gbps (port bandwidth) bare metal commodity switch and evaluated their performance and limitations. This paper also reports important findings obtained from this practical work.

Traceroute is a very useful troubleshooting tool. It reveals the routing path taken by a packet destined to a remote host. Traceroute is originally designed for the legacy IP network. However, the mechanism used by it does not work in an SDN network. In an SDN network, there has been a similar tool called SDN traceroute that provides similar functions as traceroute. However, SDN traceroute does not work in the legacy IP network. Nowadays, hybrid networks composed of both legacy and SDN switches are expected to exist for several years during the transition period. For such a hybrid network, we design and implement a tool called hybridtrace that can reveal the routing path taken by a packet that needs to traverse multiple legacy network islands and SDN network islands to reach its destination. In this paper, we present the design and implementation of hybridtrace, show the path-tracing information provided by it, and demonstrate its functions and operations.

Big-Data applications need to shuffle their data over a network in a very large amount. The ability to efficiently allocate network resources to quickly finish the shuffle greatly affects the overall processing speed of Big-Data applications. The recent emergence of software-defined networking (SDN) technology makes it easier for the network to integrate with a Big-Data application. The network can allocate network resources based on an application's run-time requirements to accelerate the speed of Big-Data delivery. In this paper, we used Hadoop as a Big-Data application platform and designed an application-aware SDN routing scheme for Hadoop to accelerate its MapReduce data shuffling over a network. Our results show that our method can outperform the ECMP-RR scheme by up to 20% and outperform the Spanning Tree method and the Floodlight OpenFlow controller by up to 70%.

Nowdays, many cloud services are provided over the OpenStack platform. How to improve the network performance of OpenStack has become an important topic. In this paper, we propose a method to mitigate network congestion in OpenStack. In our method, when network congestion occurs, a VM (virtual machine) live migration design and a flow path adjustment design are used with the SDN (software-defined networking) technology to mitigate network congestion. We performed several experiments under the fixed-rate flows and the unlimited-rate flows conditions. Then, we compared the results of our proposed method with the results achieved by using the ECMP (Equal Cost Multiple Path) routing method. We show that our proposed method reduces total file transer time in an OpenStack network and performs better than ECMP. In addition, we found that the VM live migration design is more effective in mitigating network congestion than the path adjustment design.

The SDN (Software-defined Networking) technology is a new and important technology for next-generation networks. In the current usages of SDN, the most well-known and mature protocol between an SDN controller and SDN switches is the OpenFlow protocol. OpenvSwitch is a software switch that supports the OpenFlow protocol. An OpenvSwitch is commonly run in a physical server to connect all of the virtual machines running on the server with a real network. In addition, it can run on a stand-alone Linux system to make the Linux system and its network interface cards act as a hardware switch. This paper proposes a new design and implementation for OpenvSwitch running on Linux to boost its OpenFlow control-plane message exchange performance. Our results show that our design and implementation can improve the performance by up to 58%.

Openvswitch is the most popular software implementation of an OpenFlow switch on the Linux system. Its current implementation includes a kernel module and a user space daemon. To connect with an OpenFlow controller and handle all the command message traffic, Openvswitch needs to transfer control messages between the user space and the kernel space of the Linux system frequently. When the amount of traffic becomes huge, control messages may be dropped in Openvswitch due to the lack of CPU cycles, causing performance bottlenecks. In this paper, we proposed a simple yet very effective design to improve the SDN control-plane performance of Openvswitch. The design is giving Openvswitch balanced chances to process the control messages in the netlink socket and the TCP socket so that the packets dropped in Openvswitch can be greatly reduced. Our results show that our optimized design can increase the control-plane message transfer throughput by up to 70% and make the per-second throughput very stable over time.

In this paper, we design and implement a scheme to build an optimal spanning tree over a hybrid network composed of SDN and legacy switches. SDN is an emerging network technology and gradually gaining adoption worldwide. However, during the technology transition period, which may last several years from now on, inevitably SDN switches and legacy switches will need to coexist in a network. In a legacy network with loops, the IEEE 802.11D distributed spanning tree protocol is used to construct a spanning tree to avoid the packet broadcast storm problem. In an SDN network, the SDN controller can use the collected global knowledge of the network topology to centrally construct a spanning tree to solve this problem. However, so far there is no scheme to integrate together the spanning trees separately built in multiple isolated legacy networks and isolated SDN networks to build an optimal spanning tree across the hybrid network. Our scheme presented in this paper is the first scheme designed and implemented for this purpose. In this paper, we used the EstiNet OpenFlow network simulator and emulator to simulate hybrid networks. Our scheme is implemented as a module of the Floodlight SDN controller to construct an optimal spanning tree over the simulated hybrid networks. Our simulation results show that our scheme can correctly generate an optimal spanning tree based on two performance metrics: the average path latency and the average path throughput of a spanning tree.

In this paper, we compare and evaluate the correctness, performance, and scalability of EstiNet OpenFlow simulator, EstiNet OpenFlow emulator, and Mininet OpenFlow emulator over a set of grid networks. The popular Floodlight OpenFlow controller is used without any modification to control the simulated/emulated OpenFlow switches created in these tools. We performed experiments over a set of N × N grid networks, where N = 5, 6, ..., 31 and used the real-world ping program to observe whether the average RTTs reported by these ping packets are correct or not over these tools. We found that in EstiNet simulation, the simulated results of the average RTT are always correct and repeatable, but EstiNet simulator needs more time to finish the simulation when the network size becomes larger. As for emulation, we found that Mininet emulator generated strange results that cannot be explained over some network sizes. In addition, Mininet emulator spends a huge amount of time on its program launch, network setup, and resource releasing when the network size is large. As for EstiNet emulator, we found that it generated good performance and scalability and it used less time to obtain results.

The IEEE 802.11p/1609 standards specify a suite of communication protocols for vehicular communication networks. One control channel and six service channels are used in such a network. Ideally, the data transmission load generated by all service providers and their user stations should be spread over all of the six service channels evenly to fully utilize the aggregate bandwidth of these service channels. However, the IEEE 802.11p/1609 standards do not define a way to achieve this load-balancing goal. As a result, many service providers may choose to use the same service channel and leave some service channels totally unused. This will cause heavy congestion on some service channels while wasting the bandwidth of unused service channels. To solve this problem, in this paper we propose a cooperative approach to fully utilizing the bandwidth of the six service channels in the IEEE 802.11p/1609 networks. In our approach, every service provider collects and computes the current load of the service channel that it is using and shares the computed load information with other service providers on the control channel via its WSA control packets. With this global knowledge, each service provider can choose to use an idle or lightly-used service channel to avoid heavy congestion. We used the EstiNet network simulator and emulator to conduct simulation studies. Our simulation results show that our approach significantly improves the utilization of the aggregate bandwidth of the six service channels by as high as 600%.

The Cognitive Radio (CR) technology has emerged as a hot research area recently due to its high potential in addressing the spectrum underutilization problem. In this paper, the CR is applied to unauthorized CR users (CRUs) such that these nodes can access the 802.11 spectrum opportunistically to flexibly use the detected “spectrum holes,” with the ultimate goal of improving the 802.11 spectrum utilization. In [5], the authors proposed a bi-directional CR MAC protocol such that the CRUs can smartly negotiate and reserve bandwidth either uni-directionally or bi-directionally, to better support both the TCP and UDP flows. However, the protocol design is inadequate in supporting traffic demands in more complicated network cases. To address this problem, we proposed a CR MAC protocol with a dynamic bandwidth allocation capability, which can well support various types of network traffic, particularly asymmetric two-way traffic flows. We used the NCTUns network simulator to conduct the study. Our simulation results show that the proposed protocols can further utilize the “spectrum holes” in a more effective way while guaranteeing QoS support to both the surrounding CRUs and authorized 802.11 nodes.

In this paper, we optimize the performance of a cloud platform to effectively support cooperative spectrum sensing in a cognitive radio (CR) cloud network. This cloud uses the Apache Hadoop platform to run a cooperative spectrum sensing algorithm in parallel over multiple servers in the cloud. A cooperative spectrum sensing algorithm needs to process a very large number of spectrum sensing reports per second to quickly update its database that stores the current activities of all primary users of the CR network. Because the updates of the database must be finished as soon as possible to make the CR approach effective, the cloud platform must be able to run the algorithm in real time with as little overhead as possible. In this work, we first measured the execution time of such an algorithm over our own cloud and the Amazon EC2 public cloud, using the original Hadoop platform design and implementation. We found that the original Hadoop platform has too much fixed overhead and incurs too much delay to the cooperative spectrum sensing algorithm, which makes it unable to update the primary user database in just a few seconds. Therefore, we studied the source code and the design and implementation of the Hadoop platform to improve its performance. Our experimental results show that our improvement of the Hadoop platform can significantly reduce the required time of the cooperative spectrum sensing algorithm and make it more suitable for large-scale CR networks.

This paper proposes a parallel simulation methodology to speed up network simulations on modern multicore systems. In this paper, we present the design and implementation of this approach and the performance speedups achieved under various network conditions. This methodology provides two unique and important advantages: 1) one can readily enjoy performance speedups without using an unfamiliar simulation language/library to rewrite his protocol module code for parallel simulations, and 2) one can conduct parallel simulations in the same way as when he conducts sequential simulations. We implemented this methodology and evaluated its performance speedups on the popular ns-2 network simulator. Our results show that this methodology is feasible and can provide satisfactory performance speedups under high event load conditions on wired networks.

The Cognitive Radio (CR) technology has emerged as a recent breakthrough radio technology to meet future wireless demands in terms of both spectrum efficiency and performance of new wireless applications. In an 802.11 Cognitive Radio Network (CRN), when the spectrum is not used, unauthorized users with CRs can access the frequencies assigned to authorized users in an opportunistic manner, thus dramatically improving spectrum efficiency of the network. In [5], the authors proposed a CR MAC protocol for 802.11 CRNs such that the unauthorized CR users (CRUs) can flexibly use the "spectrum holes" without affecting the performance of the authorized users. However, the protocol design proposed in [5] can degrade the throughput performance of TCP flows. Since TCP is a primary transport protocol for most of the internet applications today, it is imperative that a proposed CR MAC protocol can well support TCP flows. To address this problem, in this paper we propose a Bi- directional CR MAC (Bi-MAC) protocol such that the CRUs can smartly negotiate and reserve bandwidth based on the anticipated packet exchanges on the data channels. Our simulation results show that the Bi-MAC protocol can effectively improve both the TCP and UDP performance of CRUs and increase the network goodputs.

The Cognitive Radio (CR) technology is proposed for solving the spectrum underutilization problem and has gained great attention. By allowing unauthorized users to opportunistically access frequencies authorized to specific networks (called primary networks), the CR technology can increase the bandwidth utilization of the radio spectrum in a global perspective. In [1], the authors proposed a MAC protocol for CRs that aims to “borrow” the link bandwidth from IEEE 802.11 networks. This protocol allows CR users to opportunistically exchange data on an idle 802.11 channel. However, the protocol proposed in [1] selects an 802.11 channel for data exchange in a random manner. This random channel selection process is time-consuming for CR users to start a data transmission. In addition, the protocol does not consider the effects of hidden terminals (HTs), which often exist in wireless networks and significantly decrease the goodputs of a network. To address these problems, in this paper we propose enhancements to the protocol proposed in [1]. The proposed protocol is evaluated using extensive simulations. Our simulation results show that our proposed protocol can effectively decrease the time required for finding an idle 802.11 channel to transmit data and therefore increase the network goodputs in networks with and without the existence of HTs.

Today, Intelligent Transportation System (ITS) applications has made a great impact on improving the quality and efficiency of the transportation system. This achievement has attracted researchers to propose more ITS applications, particularly to improve the transportation safety through Vehicular Ad Hoc Networks (VANETs). The major difficulty faced by these researchers is the verification of their proposals or ideas through field-trials or real-world experiments. Therefore, VANET simulator becomes the best and the most economical alternative to verify the validity and reliability of a proposed ITS application. Among all the VANET simulators, the NCTUns 6.0 appears as one of the most powerful tool due to its integrated platform of network simulation and microscopic traffic simulation. To further enhance the capability of NCTUns 6.0, we design and implement a more realistic radio propagation model, specifically for VANETs simulation. Through extensive simulations, it is shown that the adoption of a more realistic channel model has greatly affected the vehicles' behavior, as compared to the commonly used two-ray ground channel model.

In this decade, peer-to-peer (P2P) applications have emerged as popular tools for sharing data and they have gained much attention. Such P2P applications build overlay networks on top of existing infrastructure networks and disseminate data in a peer-to-peer manner. How well the P2P overlay network maps onto the underlying physical network topology can significantly impact P2P application performances. In the literature, little work has quantitatively studied the effects of underlying physical network topologies on P2P application performance. In this paper, we used simulations to evaluate the effects of underlying physical network topologies on the performance of the real-life BitTorrent protocol and application. Our results show that underlying physical network topologies significantly affect P2P application performances.

In this paper, we design and implement a hybrid WiMAX network testbed that integrates the NCTUns network emulator with a real-life WiMAX network. This hybrid WiMAX network testbed can seamlessly connect virtual networks (emulated by NCTUns) to a real-life WiMAX network. Thus, in this testbed real-life applications on real machines can exchange data with those on emulated (virtualized) network nodes. With this unique feature, our developed testbed provides a powerful, high-fidelity, and convenient testing environment for evaluating the performances and behaviors of real applications and devices in WiMAX networks. In this paper, the architecture and methodology of the developed testbed are explained, which are general and can collaborate with any other IP-based networks. In addition, an example study on this testbed is also presented to demonstrate its usage and capabilities.

NCTUns is a novel network simulator/emulator that has many unique features over traditional network simulators/emulators. It runs on Linux and many papers/projects have used it for wireless vehicular network research. According to the NCTUns website (http://NSL.csie.nctu.edu.tw/nctuns.html), as of December 24, 2009, more than 16,164 people from 137 countries in the world have registered and downloaded this software since its first public release on November 1, 2002. In its 6.0 release, NCTUns adds new supports for large-scale microscopic wireless vehicular network (WVN) simulation. In this paper, we introduce the important capabilities of NCTUns 6.0 for wireless vehicular network research.

The IEEE 802.11(p)/1609 network is a promising candidate for future vehicular communication networks. Due to the deployment cost consideration, roadside units (RSUs) in such networks are usually installed only at hot spots and intersections, causing the service coverage of RSUs to be discontinuous. To overcome this problem, multihop data forwarding among vehicles can be used to extend the service coverage of RSUs. In this paper, we propose a receiver-centric WBSS (Wave Basic Service Set)-creating scheme to support multihop data forwarding in WBSS-based IEEE 802.11(p)/1609 networks and evaluate its performances using simulations. Our simulation results show that this scheme outperforms a typical sender-centric WBSS-creating scheme on end-to-end flow goodputs under different traffic loads.

In an IEEE 802.16(d) (WiMAX) mesh network, network bandwidth can be managed by either the centralized scheduling (CS) mode or the distributed scheduling (DS) mode. Compared with the CS mode, the DS mode provides a larger bandwidth capacity and is more scalable. However, because the DS mode uses an on-demand three-way handshake procedure to establish data schedules, the network quality experienced by application programs may fluctuate drastically. To address this problem, in this paper we propose three schemes to improve the data scheduling efficiency of the DS mode of the IEEE 802.16(d) mesh network. Our simulation results show that network performances and qualities experienced by application programs are significantly improved under our schemes.

The IEEE 802.11(p)/1609 standard is an emerging technology for vehicular communication networks. It amends the IEEE 802.11-2007 standard and defines a new WAVE operational mode for vehicular environments. The WAVE mode utilizes a combined FDMA/TDMA scheme to manage network bandwidth. In this mode, a node must periodically switch its frequency channel between two different channels over time. Such a design may result in bandwidth wastage due to the residual time caused by channel switching. In this paper, we propose two easy-to-implement schemes to mitigate this problem and compare their TCP/UDP performances with those of the original scheme. Our simulation results provide many insights into the operations of these schemes on IEEE 802.11(p)/1609 networks and show that the proposed schemes always outperform the original scheme.

NCTUns is a novel network simulator and emulator that has many unique features over traditional network simulators and emulators. It is an open-source software running on Linux and is being used by many researchers in the world. According to the NCTUns official Web site (http://NSL.csie.nctu.edu.tw/nctuns.html), as of June 1, 2008, more than 11,546 people from 124 countries have registered and downloaded this software and these numbers are still fast increasing. In its 5.0 release, NCTUns provides a complete implementation of the IEEE 802.11(p)/1609 standards defined for wireless vehicular networks. In this paper, we present the capabilities of NCTUns 5.0 focusing on its uses for IEEE 802.11(p)/1609 wireless vehicular network researches.

Road safety has become an important issue and gained much attention for many years. To avoid vehicle collisions on the roads, researchers have proposed various collision warning systems (CWS) using different technologies. Compared with CWS using other technologies, a radio-based CWS is more capable of avoiding collisions in off-sight scenarios (such as the four corners of an intersection). Most existing radio-based CWSs operate using vehicle-to-vehicle communications. However, in such CWSs, high mobility of moving vehicles easily results in message collisions and link breakage, impairing the reliability of these CWSs. An alternative is employing vehicle-to-infrastructure communication to build CWSs. In this paper, we propose a radio-based CWS employing vehicle-to-infrastructure communications. Our simulation results show that the proposed CWS can effectively reduce the number of vehicle collisions without degrading the traffic throughput of a road network.

Intelligent transportation systems (ITS) is an important research topic. One goal of ITS is to provide Internet access services for vehicle drivers and passengers. Deploying many roadside wireless access points along a highway may be able to provide such a service on the highway. However, due to the short transmission range of these wireless access points, such an approach may incur much deployment cost. This paper investigates the potential of using inter-vehicle communication to extend the coverage area of roadside access points on highways.

A DVB-RCS system provides bidirectional connections between service providers and end users. Service providers deliver broadcast data through the forward link to end users while end users send back their data through the return link. The return link is composed of multiple access channels, which can be accessed by the MF-TDMA scheme. Over each channel, a time-slot includes a guard time duration for accommodating propagation delay uncertainties among RCSTs accessing the same channel. In a system with mobile RCSTs, the guard time duration may take a considerable portion of a time-slot for avoiding data collision. Guard time represents bandwidth overhead of the system and large guard time causes significant link capacity reduction. In this paper, we first investigate the problem of link capacity reduction when the DVB-RCS standard, which is originally designed for fixed RCSTs, is directly used with mobile RCSTs without modification. Then, we propose a dynamic RCST grouping mechanism for DVB-RCS to achieve high link capacity with mobile RCSTs.

In this paper, we present an integrated simulation platform, called NCTUns, for vehicular traffic, communication, and network researches. This platform combines the capabilities supported by a network simulator and those supported by a traffic simulator. With these simulation capabilities, NCTUns can be used to design protocols for intelligent transportation systems (ITS) communication networks such as a wireless vehicular communication network. Besides, the novel architecture of the platform enables the real-world Linux protocol stack and any real-world application to be used in simulations of such networks. In this paper, we present the design of NCTUns for supporting ITS researches and show its scalability.

The IEEE 802.16 standard is a promising next-generation broadband technology. It contains two operational modes - PMP and mesh. The PMP mode defines one-hop communication between a base station (BS) and a subscriber station (SS), while the mesh mode allows traffic to be directly exchanged and forwarded among neighboring SSs. In the mesh mode, network initialization is composed of the network entry (NENT) and link establishment (LinkEst) processes. In this paper, we identify that the NENT process is prone to fail in dense networks and propose a solution to remedy this problem. Besides, we propose an extension to the LinkEst process to improve its efficiency. Both proposals are compatible with the current standard. Our simulation results show that our solution to the NENT process can on average increase the success rate by 22.5%, and our extension to the LinkEst process can on average speed up the process by 2.57 times.

Intelligent transportation systems (ITS) is an important research topic. One important function of ITS is to broadcast emergent traffic information to vehicles so that they can avoid a dangerous or congested zone in time. This paper investigates the use of inter-vehicle communication to assist roadside wireless access points to provide a lower-cost message broadcasting service on highways

Intelligent transportation systems (ITS) is an important research topic. One goal of ITS is to exchange information among vehicles in a timely and efficient manner. In the ITS research community, inter-vehicle communications (IVC) is considered a way that could achieve this goal without infrastructure support. An information network built on top of vehicles using IVC is a type of mobile ad hoc networks (MANET). In the past, many unicast routing protocols for MANET have been proposed. In these protocols, when a routing path breaks, they either locally or globally repair it. Because the topology of an IVC network changes very rapidly on the roads, an established routing path in such a network breaks very frequently. This phenomenon hurts the performance of applications running on an IVC network. In this paper, we investigate the effect of using roadside wireless repeaters on extending routing path lifetime in IVC networks on highways.

An information network built on top of vehicles using inter-vehicle communication (IVC) can be viewed as a type of mobile ad hoc network (MANET). The lifetime of the unicast routing paths existing in an IVC network can be affected by several factors such as vehicle moving speed, wireless radio transmission range, traffic density, etc. The paper uses more realistic vehicle mobility traces to study the effects of wireless transmission range on path lifetime in an IVC network. The results presented can help IVC researchers choose an appropriate wireless transmission range for IVC networks.

Abstract This paper describes NCTUns, an innovative network simulator and emulator for wireless and mobile networks. Effects of various radio resource management and quality of service (QoS) schemes on higher‐layer protocols and real‐world applications can be easily studied using NCTUns. In this paper, we elaborate on NCTUns simulation methodology, architecture, design, functionalities, performance, and applications. NCTUns simulation for wireless ad hoc, sensor, inter‐vehicle communication networks, GPRS cellular networks, and wireless mesh networks are also illustrated. More details about this tool can be found in http://NSL.csie.nctu.edu.tw/nctuns.html . Copyright © 2005 John Wiley & Sons, Ltd.

Recently, intelligent transportation systems (ITS) is becoming an important research topic. One goal of ITS is to exchange information between vehicles in a timely and efficient manner. In the ITS research community, inter-vehicle communications (IVC) is considered a way that may be able to achieve this goal. An information network built on top of vehicles using IVC can be viewed as a type of mobile ad hoc networks. Although several unicast routing protocols have been proposed for mobile ad hoc networks, how well they can operate on an IVC network is still poorly studied. If generally the lifetime of an established unicast routing path is short on an IVC network, these routing protocols will not perform well for such a network. Using several vehicle mobility traces generated by a microscopic traffic simulator, this paper analyzes the lifetime of repairable unicast routing paths existing in an IVC network. Our analytical results can help routing protocols predict the lifetime of a found path. Such a capability can lead to many useful applications.

We propose using a modified TCP decoupling approach as a congestion control mechanism for optical burst switched networks. The TCP decoupling approach is a novel way that is applicable to TCP congestion control of any traffic flow (which can be an aggregate) in a network. Since this approach is generic, it has found applications in several areas. In the optical burst switching (OBS) area, because the basic mechanism of the TCP decoupling approach matches the mechanism of the OBS very well, use of a modified TCP decoupling approach to control the traffic load handled by an OBS switch and to regulate the timing in sending bursts is proposed. Our simulation results show that this approach enables an OBS switch to achieve high link utilization while maintaining very low packet (burst) drop rate.

Mobile IP is the current standard for supporting macromobility of mobile hosts. However, in the case of micromobility support, there are several competing proposals. We present the design, implementation and performance evaluation of HAWAII (handoff-aware wireless access Internet infrastructure), a domain-based approach for supporting mobility. HAWAII uses specialized path setup schemes which install host-based forwarding entries in specific routers to support intra-domain micromobility. These path setup schemes deliver excellent performance by reducing mobility related disruption to user applications. Also, mobile hosts retain their network address while moving within the domain, simplifying quality-of-service (QoS) support. Furthermore, reliability is achieved through maintaining soft-state forwarding entries for the mobile hosts and leveraging fault detection mechanisms built in existing intra-domain routing protocols. HAWAII defaults to using Mobile IP for macromobility, thus providing a comprehensive solution for mobility support in wide-area wireless networks.

A TCP trunk is an aggregate traffic stream whose data packets are transported at a rate dynamically determined by the TCP's congestion control. Typically such a trunk is implemented on top of a layer-2 virtual circuit or an MPLS label switched path. A management TCP connection is used to regulate the rate at which the trunk transmits its data packets. Setting up a TCP trunk over a circuit or a path is easy, involving only the two end nodes of a trunk to implement the management TCP connection. A TCP trunk can guarantee minimum bandwidth while being able to grab additional bandwidth when it is available. When carried by a TCP trunk, UDP flows will be constrained in their bandwidth usage, although they themselves do not perform congestion control. Experiments on testbed networks have validated these properties. TCP trunking can be an effective tool for network operators in managing bandwidth sharing between aggregates.

Mobile-IP is the current standard for supporting macro-mobility of mobile hosts. However, in the case of micro-mobility support, there are several competing proposals. In this paper we present the design, implementation, and performance evaluation of HAWAII: a domain-based approach for supporting mobility. HAWAII uses specialized path setup schemes which install host-based forwarding entries in specific routers to support intra-domain micro-mobility. These path setup schemes deliver excellent performance by reducing mobility related disruption to user applications. Also, mobile hosts retain their network address while moving within the domain, simplifying QoS support. Furthermore, reliability is achieved through maintaining soft-state forwarding entries for the mobile hosts and leveraging fault detection mechanisms built in existing intra-domain routing protocols. HAWAII defaults to using Mobile IP for macro-mobility, thus providing a comprehensive solution for mobility support in wide-area wireless networks.

This paper proposes a simple methodology for constructing extensible and high-fidelity TCP/IP simulators in BSD UNIX environments. A simulator constructed under this methodology will simulate multiple network nodes by re-entering the UNIX kernel of the simulation host multiple times. Generated simulation results are derived from executing the native TCP/IP protocol stack on the simulation host. They are thus more accurate than those generated from a TCP/IP network simulator that implements only an abstraction of a real-life TCP/IP implementation. By using this methodology, the simulator architecture creates an illusion for the BSD UNIX kernel that the simulated network is a real network. All existing application programs such as FTP, telnet and HTTP, and all network utilities such as route, ifconfig and tcpdump are immediately applicable to a simulated network for generating network traffic, configuring networks, gathering statistics, etc. Additionally, the network simulator provides the standard UNIX API on every node in a simulated network so that ally existing or future application program can run on any node in a simulated network. This allows a network simulator to be easily extended to study high-level network architecture and application issues.

Zero queueing flow control (ZQFC) is a new credit-based flow control method for ATM networks. The receiving node of such a flow-controlled link will have zero queue-occupancy in the steady state. ZQFC uses both link and VC flow control simultaneously over the link to implement the required rate adaptation for achieving zero queueing. Because of its zero queueing property, ZQFC is able to solve a head-of-the-line blocking problem that may arise when multiple VCs share the same receiver buffer in implementing their flow control. ZQFC works well with TCP traffic. When there is any queue buildup associated with a TCP connection in a shared buffer, ZQFC will identify the connection and take steps to reduce its arrival rate at the buffer. This will allow many TCP connections to share the buffer efficiently and fairly. It makes sense to deploy ZQFC for just a single link where performance improvement is critical. Simulations using real-life TCP code have demonstrated these advantages of ZQFC.

Extensive simulation has demonstrated the effectiveness of credit-based ATM flow control in supporting client-server applications. In particular, request/response protocols used in these applications allow efficient sharing of switch buffer, for the case when all the VCs sharing the same buffer are subject to the same degree of downstream congestion. The required buffer size can be as small as the minimum of bandwidth/sup */RTT for the link and Clients/sup */Reply-Size. Request/response protocols generally tolerate congestion better than greedy loads. Being able to avoid synchronization, FIFO scheduling is sometimes more efficient than VC round-robin scheduling. These findings are derived from an online Web database of simulation results covering more than 10,000 network and load configurations. The Web database approach has proven to be effective in managing and navigating a large set of simulation results.