The Transmission Control Protocol (TCP) with Network Coding (TCP/NC) was proposed to introduce packet loss recovery ability at the sink without TCP retransmission, which is realized by proactively sending redundant combination packets encoded at the source. Although TCP/NC is expected to mitigate the goodput degradation of TCP over lossy networks, the original TCP/NC does not work well in burst loss and time-varying channels. No apparent scheme was provided to decide and change the network coding-related parameters (NC parameters) to suit the diverse and changeable loss conditions. In this paper, a solution to support TCP/NC in adapting to mentioned conditions is proposed, called TCP/NC with Loss Rate and Loss Burstiness Estimation (TCP/NCwLRLBE). Both the packet loss rate and burstiness are estimated by observing transmitted packets to adapt to burst loss channels. Appropriate NC parameters are calculated from the estimated probability of successful recoverable transmission based on a mathematical model of packet losses. Moreover, a new mechanism for coding window handling is developed to update NC parameters in the coding system promptly. The proposed scheme is implemented and validated in Network Simulator 3 with two different types of burst loss model. The results suggest the potential of TCP/NCwLRLBE to mitigate the TCP goodput degradation in both the random loss and burst loss channels with the time-varying conditions.

A number of studies have been conducted on topic modeling for various types of data, including text and image data. We focus particularly on the burstiness of the local features in modeling topics within video data in this paper. Burstiness is a phenomenon that is often discussed for text data. The idea is that if a word is used once in a document, it is more likely to be used again within the document. It is also observed in video data; for example, an object or visual word in video data is more likely to appear repeatedly within the same video data. Based on the idea mentioned above, we propose a new topic model, the Correspondence Dirichlet Compound Multinomial LDA (Corr-DCMLDA), which takes into account the burstiness of the local features in video data. The unknown parameters and latent variables in the model are estimated by conducting a collapsed Gibbs sampling and the hyperparameters are estimated by focusing on the fixed-point iterations. We demonstrate through experimentation on the genre classification of social video data that our model works more effectively than several baselines.

Current literature on input buffer management reveals that, in representative ATM networks under highly bursty traffic conditions, the fuzzy thresholding approach yields lower cell loss rate at the cost of lower throughput. Also, under less bursty traffic, the traditional fixed thresholding approach achieves higher throughput at the expense of higher cell loss rate. The integration of these two properties into practice is termed adaptive dynamic buffer management (ADBM) approach for input buffers and its assessment is the objective of this paper. The argument is that, given that the traffic conditions are constantly changing, to achieve efficiency during actual operation, the network control must dynamically switch, at every ATM switch, under the call processor's control, between the two input buffer management techniques, dictated by the nature of the traffic at the inputs of the corresponding switch. The need to involve the call processor marks the first effort in the literature to dynamically configure input buffer management architectures at the switch fabric level under higher level call processor control. It stems from the fact that the switch fabric operates very fast and cannot engage in complex decision making without incurring stiff penalty. To achieve this goal, the network control needs knowledge of the burstiness of the traffic at the inputs of every ATM switch. The difficulties with this need are two-fold. First, it is not always easy to obtain the traffic model and model parameters for a specific user's call. Second, even where the traffic model and the model parameters are known for a specific user's call, this knowledge is valid only at the source switch where the user interfaces with the network. At all other switches in the network, the cells of the traffic in question interact asynchronously with the cells from other traffic sources and are subject to statistical multiplexing. Thus, to obtain the exact nature of the composite traffic at the inputs of any ATM switch, is a challenge. Conceivably, one may determine the burstiness by counting the number of cells incurred at the inputs of an ATM switch over a defined time interval. The challenge posed by this proposition lies in the very definition of burstiness in that the time interval must approach, in the limit, zero or the resolution of time in the network. To address this challenge, first, a 15-node representative ATM network is modeled in an asynchronous, distributed simulator and, second, simulated on a network of workstations under realistic traffic stimuli. Third, burstiness indices are measured for the synthetic, stochastic traffic at the inputs of every ATM switch as a function of the progress of simulation for different choices of time interval values, ranging from 20,000 timesteps down to 1,000 timesteps. A timestep equals 2.73 µs. Results reveal that consistent burstiness indices are obtained for interval choices between 1,000 and 5,000 timesteps and that a burstiness index of 25, measured at 3,000 timestep interval, constitutes a reasonable and practical threshold value that distinguishes highly bursty traffic that warrants the use of the fuzzy thresholding approach from less bursty traffic that can benefit from the fixed thresholding scheme. A comparative performance analysis of ADBM yields the following. For pure fixed and pure fuzzy thresholding schemes, the throughputs are at 73.88% and 71.53% while the cell drop rates are at 4.31% and 2.44%,respectively. For the ADBM approach, where the input buffer management alternates at each individual ATM switch between the fixed and fuzzy schemes, governed by measured burstiness index threshold of 25 for a 3,000 timestep interval, the throughput is 74.77%, which is higher than even the pure fixed scheme while the cell drop rate is 2.21% that is lower than that of the pure fuzzy scheme. In essence, ADBM successfully integrates the best characteristics of the fuzzy and fixed thresholding schemes.

A principal attraction of ATM networks, in both wired and wireless realizations, is that the key quality of service (QoS) parameters of every call, including end-to-end delay, jitter, and loss are guaranteed by the network when appropriate cell-level traffic controls are imposed at the user network interface (UNI) on a per call basis, utilizing the peak cell rate (PCR) and the sustainable cell rate (SCR) values for the multimedia--voice, video, and data, traffic sources. There are three practical difficulties with these guarantees. First, while PCR and SCR values are, in general, difficult to obtain for traffic sources, the typical user-provided parameter is a combination of the PCR, SCR, and the maximum burstiness over the entire duration of the traffic. Second, the difficulty in accurately defining PCR arises from the requirement that the smallest time interval must be specified over which the PCR is computed which, in the limit, will approach zero or the network's resolution of time. Third, the literature does not contain any reference to a scientific principle underlying these guarantees. Under these circumstances, the issue of providing QoS guarantees in the real world, through traffic controls applied on a per call basis, is rendered uncertain. This paper adopts a radically different, high level approach to the issue of QoS guarantees. It aims at uncovering through systematic experimentation a relationship, if any exists, between the key high level user traffic characteristics and the resulting QoS measures in a realistic operational environment. It may be observed that while each user is solely interested in the QoS of his/her own traffic, the network provider cares for two factors: (1) Maximize the link utilization in the network since links constitute a significant investment, and (2) ensure the QoS guarantees for every user traffic, thereby maintaining customer satisfaction. Based on the observations, this paper proposes a two-phase strategy. Under the first phase, the average "link utilization" computed over all the links in a network is maintained within a range, specified by the underlying network provider, through high level call admission control, i.e. by limiting the volume of the incident traffic on the network, at any time. The second phase is based on the hypothesis that the number of traffic sources, their nature--audio, video, or data, and the bandwidth distribution of the source traffic, admitted subject to a specific chosen value of "link utilization" in the network, will exert a unique influence on the cumulative delay distribution at the buffers of the representative nodes and, hence, on the QoS guarantees of each call. The underlying thinking is as follows. The cumulative buffer delay distribution, at any given node and at any time instant, will clearly reflect the cumulative effect of the traffic distributions of the multiple connections that are currently active on the input links. Any bounds imposed on the cumulative buffer delay distribution at the nodes of the network will also dominate the QoS bounds of each of the constituent user traffic. Thus, for each individual traffic source, the buffer delay distributions at the nodes of the network, obtained for different traffic distributions, may serve as its QoS measure. If the hypothesis is proven true, in essence, the number of traffic sources and their bandwidth distribution will serve asa practically realizable high level traffic control in providing realistic QoS guarantees for every call. To verify the correctness of the hypothesis, an experiment is designed that consists of a representative ATM network, traffic sources that are characterized through representative and realistic user-provided parameters, and a given set of input traffic volumes appropriate for a network provider approved link utilization measure. The key source traffic parameters include the number of sources that are incident on the network and the constituent links at any given time, the bandwidth requirement of the sources, and their nature. For each call, the constituent cells are generated stochastically, utilizing the typical user-provided parameter as an estimate of the bandwidth requirement. Extensive simulations reveal that, for a given link utilization level held uniform throughout the network, while the QoS metrics--end-to-end cell delay, jitter, and loss, are superior in the presence of many calls each with low bandwidth requirement, they are significantly worse when the network carries fewer calls of very high bandwidths. The findings demonstrate the feasibility of guaranteeing QoS for each and every call through high level traffic controls. As for practicality, call durations are relatively long, ranging from ms to even minutes, thereby enabling network management to exercise realistic controls over them, even in a geographically widely dispersed ATM network. In contrast, current traffic controls that act on ATM cells at the UNI face formidable challenge from high bandwidth traffic where cell lifetimes may be extremely short, in the range of µs. The findings also underscore two additional important contributions of this paper. First, the network provider may collect data on the high level user traffic characteristics, compute the corresponding average link utilization in the network, and measure the cumulative buffer delay distributions at the nodes, in an operational network. The provider may then determine, based on all relevant criteria, a range of input and system parameters over which the network may be permitted to operate, the intersection of all of which may yield a realistic network operating point (NOP). During subsequent operation of the network, the network provider may guide and maintain the network at a desired NOP by exercising control over the input and system parameters including link utilization, call admittance based on the requested bandwidth, etc. Second, the finding constitutes a vulnerability of ATM networks which a perpetrator may exploit to launch a performance attack.

Real traffic flows are captured in various network environments and their statistical properties are analyzed. Based on real traffic flows, MWM (Multifractal Wavelet Model) and Poisson equivalent synthetic traffic flows are generated. Performance analysis of a SB (Synchronous Bulk) packet switch is joined with different types of traffic. Maximum throughput performance of the SB packet switch for various real traffic flows and appropriate MWM and Poisson equivalent synthetic traffic flows are evaluated by using discrete-event simulations. Different flow persistence, SF (Stretch Factor) and scheduling mechanisms are used in order to asses their influence on SB packet switch performance. Traffic asymmetry, either input or output based, has a major influence on SB packet switch performance. By increasing the level of asymmetry, maximum throughput values decrease considerably, especially if the ROT (Rotation) scheduling mechanism is applied. Traffic asymmetry also decreases the influence of the SF parameter on maximum switch throughput. As a general rule of thumb, SF values of no more then 5 must be used if asymmetrical traffic is switched. It is also advisable that OPF (Oldest Packet First) scheduling mechanism is used in such cases. The influence of burstiness and scaling of traffic flows turns out to be relatively insignificant for the SB packet switch maximum throughput results, if the OPF scheduling mechanism is used. Larger throughput discrepancies are detected, if ROT scheduling is used.

A per-connection end-to-end call admission control (CAC) problem is solved in this paper to allocate network resources to an input session to guarantee its quality of service (Qos) requirements. In conjunction with the solution of the CAC problem, a traffic descriptor is proposed to describe the loss rate and the delay bound Qos requirements of the connection to be set up as well as the statistical characteristics of the associated input traffic which is modeled as a linear mean function plus a (zero-mean) fractional Brownian motion. The information in the traffic descriptor is sufficient to determine the allocation of channel bandwidth and buffer space to the input traffic in a network which employs leaky bucket shapers and scheduling algorithms to guarantee the Qos requirements. The CAC problem is solved by an iterative algorithm of which there are two stages in each iteration: one is responsible for the search of a candidate end-to-end routing path and the other for the verification of the legitimacy of this candidate path to meet the Qos requirements and for the allocation of resources in such a legitimate path.

In this paper, we present a traffic scheduling algorithm, called the Delay-Bound Monotonic with Average Rate Reservation (DM/ARR), which generates minimum output burstiness streams. We assume that connection i is policed by the leaky bucket algorithm with parameters (σi,ρi) where σi is the bucket size (or burstiness) and ρi is the leaky rate. Compared with the totally isolated scheme where connection i is allocated a bandwidth ri=max{σi/di,ρi} (di is the delay bound requirement of connection i), the DM/ARR algorithm has a better performance in the sense that it has a larger admission region. We prove that, among all possible scheduling algorithms that satisfy the delay bound requirements of established connections, DM/ARR results in the minimum output burstiness. This is important because a smaller burstiness implies a smoother traffic and thus the receiver (or next switch node in a multihop network) can handle it more easily. Numerical results show that the admission region of the DM/ARR algorithm is close to that of the earliest deadline first algorithm. A packetized version is studied for ATM networks.

From our previous studies, we derived the worst case cell delay within an ATM switch and thus can find the worst case end-to-end delay for a set of real-time connections. We observed that these delays are sensitive to the priority assignment of the connections. With a better priority assignment scheme within the switch, the worst case delay can be reduced and provide a better network performance. We extend our previous work on the closed form analysis to conduct more experimental study of how different priority assignments and system parameters may affect the performance. Furthermore, from our worst case delay analysis on a regulated ATM switch, network traffic can be smoothed by a leaky bucket at the output controller for each connection. With the appropriate setting on the leaky bucket parameter, the burstiness of the network traffic can be reduced without increasing the delay in the switch. Therefore, fewer buffers will be required for each active connection within the switch. In this paper, our experimental results have shown that the buffer requirement can be reduced up to 5.75% for each connection, which could be significant, when hundreds of connections are passing through the switches within a regulated ATM network.

Weighted Round Robin (WRR) scheduling is an extension of round robin scheduling. Because of its simplicity and bandwidth guarantee, WRR cell scheduling is commonly used in ATM switches. However, since cells in individual queues are sent cyclically, the delay bounds in WRR scheduling grow as the number of queues increases. Thus, static priority scheduling is often used with WRR to improve the delay bounds of real-time queues. In this paper, we show that the burstiness generated in the network is an even greater factor affecting the degradation of delay bounds. In ATM switches with per-class queueing, a number of connections are multiplexed into one class-queue. The multiplexed traffic will have a burstiness even if each connection has no burstiness, and when the multiplexed traffic is separated at the down stream switches, the separated traffic will have a burstiness even if the multiplexed traffic has been shaped in the upstream switches. In this paper, we propose a new WRR scheme, namely, WRR with Save and Borrow (WRR/SB), that helps improving the delay bound performance of WRR by taking into account the burstiness generated in the network. We analyze these cell scheduling methods to discuss their delay characteristics. Through some numerical examples, we show that delay bounds in WRR are mainly dominated by the burstiness of input traffic and, thus WRR/SP, which is a combination of WRR and static priority scheduling, is less effective in improving delay bounds. We show that WRR/SB can provide better delay bounds than WRR and that it can achieve the same target delay bound with a smaller extra bandwidth, while large extra bandwidth must be allocated for WRR.