Data centers connect a larger number of servers requiring IO and switches with low power and delay. Virtualization of IO and network is crucial for these servers, which run virtual processes for computing, storage, and applications. We propose using the PCI Express (PCIe) protocol and a new PCIe switch fabric for IO and switch virtualization. The switch fabric has little data buffering, allowing up to 512 physical 10 Gb/s PCIe 2.0 lanes to be connected via a switch fabric on a chip. The switch is scalable with adapters running multiple adaptation protocols, such as Ethernet over PCIe, PCIe over Internet, or FibreChannel over Ethernet. Such adaptation protocols allow integration of IO often required for disjoint data center applications such as storage and networking. The novel switch fabric based on space-time carrier sensing facilitates high bandwidth, low power, and low delay multiprotocol switching. To achieve Terabit switching, both time (high transmission speed) and space (multi-stage interconnection network) technologies are required. In this paper, we present the design of an up to 256 lanes Close-network of multistage crossbar switch fabric for PCIe system. The switch core consists of 32 16x16 crossbar sub-switches. We also propose a new output contention resolution algorithm utilizing an out-of-band protocol of Request-To-Send (RTS), Clear-To-Send (CTS) before sending PCIe packets through the switch fabric. Preliminary power and delay estimates are provided.

To achieve Terabit and Petabit switching, both time (high transmission speed) and space (multi-stage interconnection network) technologies are required. We propose an Ethernet for which for both time and space are carrier sensed. We extend CSMA/CD to a time- space protocol called CSMA/TS. We call the space-time transmission medium Terabit Ethernet (TbE). This space sensing CSMA/TS protocol allows routing in ether to be done by the network interface card (NIC). The advantages are scalability, lower cost, and most importantly, reduced delay end-to-end. Simple analysis is given for evaluating throughput for 2-stage and 3-stage TbE networks.

We propose a companion solution to iSCSI that is more suited for virtualization of local computer bus architectures, such as PCI/PCI-X and PCI Express. We explore the architecture of a possible virtualization protocol. The solution would allow extension of the host computer bus via an IP network based on 10Gbps Ethernet. This solution would allow adapter cards to be located remote from host systems via the Internet, yet appear as a local bus resource. We present Quality of Service (QOS) goals for PCI Express virtualization and expansion via IP. Critical QOS characteristics such as data transfer rates and latency are discussed. PCI Express transaction types are reviewed to assess whether IP over 10 Gbps Ethernet is suitable as a transport. A general protocol architecture as presently envisioned is described. We introduce a class system for performance categorization. We identify the areas that require further research and investigation. System overhead, latency, and timeouts must be addressed. Methods for configuring and optimizing the data transfer path must be developed.

IP storage area networks are based on bulk data transfer over a binary erasure channel and associated bulk data storage in end systems, which are also subjected to erasures. The paper motivates a possibility of an integrated approach to transport and data storage reliability in storage area networks (SAN) which reduces the protocol processing complexities. The integrated transport/storage reliability is sought through the novel Ying-Yang code and the convolution code with new decoding techniques.

A new architecture for storage area networks (SANs) is proposed for providing efficient information processing and transportation. We deviate from the byte stream oriented transmission control protocol (TCP) transport mechanism to more storage friendly block oriented transport. Data is processed, encrypted, error checked, redundantly encoded, and stored in fixed size blocks called quanta. Each quantum is processed by an effective cross-layer protocol that collapses the protocol stack for security, iWARP and iSCSI functions, transport control, and even redundant arrays of inexpensive disk (RAID) storage. This streamlining produces a highly efficient protocol with fewer memory copies and places most of the computational burden and security safeguard on the client, while the target stores quanta from many clients with minimal processing. We propose a new network RAID storage method using the quantum concept. Also, we unify error control and flow control of the iSCSI and TCP protocols in a manner we believe is more suitable for high data rate and low latency storage applications.

In this paper, a novel method is proposed to do turbo coding for pulse (pattern) position modulation, for ultra wideband radio system. Such systems maps multiple PPM slot detections into the probabilities of those bits to be "1", then turbo decoding can be done. The noise model in this coding scheme and the discrete interference model to replace the AWGN model to improve the coding performance are discussed.

Abstract We design a mobile ad‐hoc network (MANET), where nodes are equipped with several multi‐beam adaptive array (MBAA) systems. Such devices provide a great advantage for both space and frequency freedom, and produce a much better network throughput than traditional systems based on single frequency, omnidirectional transmission. A license‐free bandwidth up to 7 GHz is available centered at 60 GHz. The problem to divide this bandwidth into several channels is described as frequency division issue in this paper. We have provided a possible solution for such a problem in our paper, formulated as a (K,M) list‐coloring problem of the associated graph. Major results are presented based on two different models for connection requests. In addition, a numerical test is performed to verify the analysis. Copyright © 2004 John Wiley & Sons, Ltd.

We present a scalable distance-matrix routing algorithm for clusters of nodes. These clusters are defined by GPS coordinates similar to hierarchical routing and geography based routing. We show that the computation and storage requirement of this distance-matrix routing algorithm is scalable and is substantially reduced for large number of mobile nodes. Optimality of routing is maintained. The effect of geographic proximity is shown to reduce computation, message exchange, and route storage requirements. Also, the reduced number of iterations required also improves on the stability of Bellman-Ford type distributed routing algorithms. We plan to implement an IP routing protocol using existing IP router software such as RIP, allowing such mature software to scale up to much larger number of nodes, to handle node mobility, and to perform many ad-hoc network features.

In this paper, we consider the problem of designing a topology for deploying a free space optical (FSO) link based network. The problem is to create a topology with strong connectivity and short diameter with uniform degree bounds on each node. Two centralized approaches are presented. The first approach constructs a backbone network by Delaunay triangulation. The basic structure is then refined to meet the design objectives. The second approach called the closest neighbor (CN) algorithm constructs a degree constrained minimum weight spanning tree. The tree is developed into a network with good connectivity and small diameter by forming edges with the closest neighbors. We prove that the CN algorithm forms a connected network. Through simulation and analysis we also show that this approach results in high reliability and small diameter.

A hybrid fast restoration scheme for high-speed laser networks is introduced. Distributed restoration is achieved by a special restoration strategy, "local recovery for local failure". Implementation over the OSPF protocol is presented, which consists of preplanned and real-time scenarios. Recovery time is also given. Simulation results over a typical network structure show that we could achieve a full recovery of a single-link failure, 99.16% recovery of single-node failure, as well as 96.64% recovery of double-link failure.

Pulse radio has been considered a power efficient method for communication in a multiple path fading environment. We extend the method to the use of multiple transmitters per user, and the use of receiver arrays distributed over a geographic area. We show that we can obtain a linear growth in transmission capacity if we use multiple receivers. We explore the notion of spatial capacity, similar to the earlier results for point-to-point communications with multiple antenna elements (MAE). Since spatially separated users each uses a different set of paths to reach a receiver, we can imagine that the multipath "impairment" actually could be used to our advantage to improve capacity and reduce transmission errors. In effect, each user has a multipath transmission signature that helps to distinguish its information transmission. This transmission signature becomes more complex when we have multiple receivers at the base station, and multiple transmitters at the user. The signature can be seen as a spatial CDMA code, and the use of MAE actually increases the multiple access capacity linearly with the multiplicity of antennas. We study the spectral characteristics of monocycle pulses and other bandwidth efficient pulse shapes. We propose a novel technique of using a convolution formula for the Poisson shot noise process to calculate the probability distribution of multiple access interference (MAI). The technique analyses the Poisson process. Using the distribution, we calculate the channel capacity of spatial CDMA for various pulse shapes. We show that spatial CDMA provides high capacity wireless multiaccess communications.

We study the problem of real time VBR video traffic smoothing. One fundamental difficulty in this problem is that at any time during transmission, the information of most future video frames is not available and thus it is unlikely that global optimization can be achieved in the smoothing process. In order to measure the effectiveness of real time video smoothing methods, we first propose a benchmark algorithm which achieves optimality on some of the smoothness parameters in the smoothing results. Based on this algorithm, we found that significant discrepancy exists between the results produced by some of the existing smoothing methods and the smoothness upper bounds. With this observation, we then focus on devising an algorithm which improves the smoothing results. Experimental results show that our algorithm makes noticeable improvements in some of the smoothness parameters compared to existing smoothing methods.

Broadband network are developed to carry a wide range of traffic which has different characteristics and grade-of-service (GoS) requirements. To meet these distinct requirements, we propose to decompose each network component into multiple sub-components. Each sub-component has a dedicated bandwidth and buffer, and it only carries traffic which has a similar GoS requirement. Thus it will reduce the interference from other types of traffic which has totally different GoS requirements. First, we develop a GoS-based pricing scheme which will entail the resources are efficiently utilized. The equilibrium stability is also studied. Then we address the problem of how to optimally allocate available resources among the sub-components. Resource expansion becomes necessary when the network is congested. Two kinds of routing schemes are proposed. One is least cost routing for GoS-insensitive traffic and the other is best GoS routing for GoS-sensitive traffic.

We consider the use of large buffers and feedback as a mechanism to maintain loosely coupled synchronization between a multimedia server and a client. The multimedia stream is modeled as a fluid flow through rate controlled valves and buffers with multiple thresholds. These thresholds are used to control the rates upstream. The quality of service for the multimedia connection is characterized in terms of the jitter in the received media stream due to buffer underflow and overflow. This quality of service is used to exercise rate and admission control in the presence of congestion. The feedback mechanism is, implemented in GRAMS (gopher-style real time ATM multimedia system), an adaptive multimedia client-server system. Experimental statistics are gathered for the purpose of traffic engineering. We employ a fluid flow and first passage time analysis to understand the traffic process through the pipelines and the buffers and to estimate the amount of signaling required by the feedback mechanism.

A scheme for delivery or variable bit-rate (VBR) video over asynchronous transfer mode (ATM) networks where bandwidth can be renegotiated during the duration of a call between the video source and the network is considered. Renegotiation can be initiated by either the video source or the network. The video bandwidth requirement is characterized by a usage parameter control (UPC) consisting, in general, of peak rate, burst length, and sustained rate. A baseline design is outlined where rate-control adjusts the source's rate while a new UPC is requested from the network. When granted, the new UPC allows the source to maintain its target quantization and delay requirements. Rate control epochs may be extended when the network blocks UPC requests or sets a lower UPC value to temporally deal with congestion. Simulation results are presented for VBR MPEG video. The results show that with a moderate renegotiation rate the scheme tracks the bandwidth requirements of the source. As a result, the video quality and bandwidth efficiency can be maintained.

In Part I, we proposed a distributed, state dependent, dynamic routing method, called least cost routing in multi-service networks (LCRM), for multirate circuit-switched broadband ISDN networks, which support a variety of traffic classes with different characteristics. This method represents the one-step policy improvement on a chosen base policy. As a base policy /spl pi//sub 0/, we use the policy which allows direct link routing only. The policy /spl pi//sub 0/ also requires that each call class has a portion of link bandwidth dedicated to it. The task is to find an efficient computational way to obtain the corresponding relative cost values, which represent costs of routing a call over one link. We consider the one-link model with two types of traffic. Our main result is that the system of linear equations associated with the two-dimensional Markov chain can be decomposed into several systems of linear equations, each one associated with the one-dimensional Markov chain.

We propose a thermodynamic theory for broadband networks relating quantities such as grade of service (GoS), bandwidth assignment, buffer assignment, and bandwidth demand. We propose a scalability postulate for these four quantities. Useful thermodynamic type relationships are then derived. The scalability postulate and thermodynamic type relations are then reexamined via statistical methods using the moment generating function. Large deviations theory is applied, and in the process, notions such as effective bandwidth are defined. We apply this theory to networks which allow dynamic routing of different call types. The probability of rare events expressed in conjunctive forms is characterized using large deviations theory. Based on this theory, a new dynamic routing method called effective bandwidth network routing (EBNR) is proposed.>

Gopher-style real-time ATM multimedia services (GRAMS) is a multimedia system designed for an ATM local area network with the server providing multimedia services to multiple users in real-time. The major goal of this system is to serve heterogeneous media requests so that every client served has satisfactory service while the network bandwidth and the host capacity are fully exploited. Therefore, a server that can determine the service rates and successfully multiplex media transmissions according to the quality-of-service (QoS) requirement of each request becomes the key component of this system. However, given the heterogeneity of QoS requirements from different types of media, it is usually very difficult to clearly specify and accurately measure the QoS for these media. Also, it is difficult to control the QoS in a distributed-multimedia environment. We propose in this paper a simple yet effective way to measure QoS in the GRAMS system. Based on the GRAMS system architecture, we use different kinds of starvation counters to measure system resource utilization and individual service quality. These counters are employed as major QoS parameters for the admission and rate-control algorithms in the GRAMS server that are under intensive evaluation and refinement. Experimental results are also presented, showing the feasibility of simultaneously providing tens of video and images transfer sessions using low-cost workstation technology. Bottlenecks for the video server are identified.>

The problem of dynamic routing in multiple-service networks is studied. The authors consider a multirate circuit-switched broadband ISDN network that supports a variety of traffic classes with different traffic characteristics (bandwidth requirement, call arrival rate, and call holding time) and reward parameter. The task is to find a dynamic routing method which minimizes the long-run average cost of lost calls. Markov decision theory, with the objective of an average cost optimization is appropriate to the solution of this problem. Using this approach, they propose a distributed, state dependent, dynamic routing method which they call least cost routing in multiple-service networks (LCRM). They define a base policy such that the corresponding relative cost function associated with the Howard (1960) policy improvement routine of Markov decision theory is easily computed. This function is then used to determine the one-step policy improvement (LCRM).>

S.723-731

Bandwidth is engineered for heterogeneous service via a layered notion of required bandwidth. The required bandwidth is requested to satisfy grade of service up to the layer of concern. The most important use of this layer required bandwidth is for bandwidth allocation at the path layer. The mathematical properties of the optimal policy are explored. Based on these properties, a fast and accurate algorithm is derived for computing required bandwidth. Numerical examples are given.>

The authors describe a multilayer connection control architecture for broadband communications. A graph framework is introduced to describe network layers of network design, path configurations, dynamic call routing, burst switching, and asynchronous transfer mode (ATM) cell switching. These hierarchical layers of switching are performed at decreasing time scales, respectively. Switching at the higher layer is performed to reduce blocking at the next smaller time scale. A layered notion of equivalent bandwidth for satisfying layered grade-of-service parameters is introduced for making connections at these time scales. The authors then focus on the path configuration layer. Two path setup methods, namely, physical and virtual path setup, are described. Mathematical programs minimizing path bandwidth usage subject to meeting grade-of-service requirements are formulated for both methods. The relative merits of these methods are compared. In one example, physical path setup is shown to require roughly 50% more bandwidth than virtual path setup.>

The major benefit of a broadband integrated ATM (asynchronous transfer mode) network is flexible and efficient allocation of communications bandwidth for communications services. However, methods are needed for evaluating congestion for integrated traffic. The author suggests evaluating congestion at different levels, namely the packet level, the burst level, and the call level. Congestion is measured by the probabilities of packet blocking, burst blocking, and call blocking. He outlines the methodologies for comparing these blocking probabilities. The author uses the congestion measures for a multilayer bandwidth-allocation algorithm, emulating some function of virtual circuit setup, fast circuit switching, and fast packet switching at these levels. The analysis also sheds insight on traffic engineering issues such as appropriate link load, traffic integration, trunk group and switch sizing, and bandwidth reservation criteria for two bursty services.>

This paper gives a broadband (total throughput approaching 1 terabit/s) self-routing packet switch design for providing flexible multiple bit-rate broadband services for an end-to-end fiber network. The switch fabric for the slotted broadband packet switch delivers exactly one packet to each output port from one of the input ports which request packet delivery to that output port. The denied requests would try again during the next slot. We discover an effective scheme, implemented by CMOS VLSI with manageable complexity, for performing this function. First, each input port sends a request for a port destination through a Batcher Sorting network, which sorts the request destinations in ascending order so that we may easily purge all but one request for the same destination. The winning request acknowledges its originating port from the output of the Batcher network, with the acknowledgment routed through a Batcher-banyan selfrouting switch. The acknowledged input port then sends the full packet through the same Batcher-banyan switch without any conflict. Unacknowledged ports buffer the blocked packet for reentry in the next cycle. We also give several variations for significantly improved performance. We then study switch performance based on some rudimentary protocols for traffic control. For the basic scheme, we analyze the throughput-delay characteristics for random traffic, modeled by random output port requests and a binomial distribution of packet arrival. We demonstrate with a buffer size of around 20 packets, we can achieve a 50 percent loading with almost no buffer overflow. Maximum throughput of switch is 58 percent. Next, we investigate the performance of the switch in the presence of periodic broadband traffic. We then apply circuit switching techniques and packet priority for high bit-rate services in our packet switch environment. We improve the throughput per port to close to 100 percent by means of parallel switch fabric, while maintaining the periodic nature of the traffic.

We consider the use of multiple high-capacity fibers for communications networking. Each user transmits, asynchronously, patterns of optical pulses distributed over the fibers and throughout a time frame. Each receiver has a distinct alphabet of patterns, which are detected by optical correlators. Optical correlation by fiber tapped delay lines provides speedy and easy-to-implement decoders. Thus the individual user obtains a transparent low-speed channel by code multiplexing. The reliability of this low-speed channel can be enhanced by redundantly coding the patterns sent by the user, for which the encoding and decoding processes can be performed electronically. This two-step encoding process is simple to implement, highly reliable at reasonable throughput, and provides asynchronous access with simple protocol. Various components and configurations of this access scheme are described. The information theoretic capacity and the error probability for these configurations are derived. We also demonstrate that hardlimiting and filtering at the receiver reduce error probability significantly.