Journal of High Speed Networks

This work presents a novel feedback rate regulator using the multiple leaky bucket (MLB) for variable bit rate (VBR) self-similar traffic that is based on the traffic load prediction by time-delayed neural networks in ATM networks. In the MLB mechanism, the leak rate and buffer capacity of each leaky bucket (LB) can be dynamically adjusted based on the buffer occupancy. A finite-duration impulse response (FIR) multilayer neural network is used to predict the incoming traffic load and pass the information to the feedback rate regulator. Ten real world MPEG1 and ten synthesized traffic traces are used to validate the performance of the MLB and the MLB with an FIR prediction mechanism. Simulation results demonstrate that the cell loss rate using MLB and MLB with an FIR filter-based predictor can be significantly reduced compare to the conventional leaky bucket method.

Shortest Path Tree (SPT) is the most widely used type of tree for multicast provisioning due to its simplicity and low per-destination cost. An SPT minimizes the accumulated cost, individually, from the source of a group to each destination of the group. However, SPTs have not considered the overall resource utilization in their constructions. This work aims at building cost-effective SPTs by enhancing link sharing between destinations of a group. We achieve this goal by introducing destination-driven characteristic into SPT constructions. Specifically, each destination is connected with the source via a shortest path. When equal cost multiple paths are available, priority is given to the one biasing through a destination among all such routes. We accordingly present the design of an algorithm building destination-driven SPTs. To achieve further improved performance in resource utilization, we also present an algorithm, which is designed to further enhance link sharing among the destinations of a group while meeting a maximum path length constraint for each destination. Simulation results are used to demonstrate the high performance of the proposed algorithms.

This paper presents a novel architecture of internally two priority buffered Multistage Interconnection Network (MIN). First, we compare by simulation the new architecture against a single priority MIN and demonstrate up to N times higher throughput for the high priority traffic in a hot spot situation, when N is the number of inputs. In addition, under uniform traffic assumption we show an increase in the low priority throughput, without any change in the high priority throughput. Moreover, while in the single priority system the high priority delay and its standard deviation are increased when low priority traffic is present, it is kept constant in the dual priority system. Finally, we introduce a new approach of long Markovian memory performance model to better capture the packets dependency in a single priority MIN under uniform traffic and extend this model for a dual priority MIN. Model results are shown to be very accurate.

Interrupt processing can be a major bottleneck in the end-to-end performance of high-speed networks. The performance of Gigabit network end hosts or servers can be severely degraded due to interrupt overhead caused by heavy incoming traffic. Under heavy network traffic, the system performance will be negatively affected due to interrupt overhead caused by the incoming traffic. In particular, excessive latency and significant degradation in system throughput can be experienced. In this paper, we present a throughput-delay analysis of such behavior. We develop analytical models based on queueing theory and Markov processes. In our analysis, we consider and model three systems: ideal, PIO, and DMA. In ideal system, the interrupt overhead is ignored. In PIO, DMA is disabled and copying of incoming packets is performed by the CPU. In DMA, copying of incoming packet is performed by DMA engines. For high-speed network hosts, both PIO and DMA can be desirable configuration options. The analysis yields insight into understanding and predicting the impact of system and network choices on the performance of interrupt-driven systems when subjected to light and heavy network loads. Simulations and reported experimental results show that our analytical models are valid and give a good approximation.

IPsec is the current security standard for the Internet Protocol IP. According to this standard, a selected computer pair (p, q) in the Internet can be designated a “security association”. This designation guarantees that all sent IP messages whose original source is computer p and whose ultimate destination is computer q cannot be replayed in the future (by an adversary between p and q) and still be received by computer q as fresh messages from p. This guarantee is provided by adding increasing sequence numbers to all IP messages sent from p to q. Thus, p needs to always remember the sequence number of the last sent message, and q needs to always remember the sequence number of the last received message. Unfortunately, when computer p or q is reset these sequence numbers can be forgotten, and this leads to two bad possibilities: unbounded number of fresh messages from p can be discarded by q, and unbounded number of replayed messages can be accepted by q. In this paper, we propose two operations, “SAVE” and “FETCH”, to prevent these possibilities. The SAVE operation can be used to store the last sent sequence number in persistent memory of p once every K_p sent messages, and can be used to store the last received sequence number in persistent memory of q once every K_q received messages. The FETCH operation can be used to fetch the last stored sequence number for a computer when that computer wakes up after a reset. We show that the following three conditions hold when SAVE and FETCH are adopted in both p and q. First, when p is reset, at most 2K_p sequence numbers will be lost but no fresh message sent from p to q will be discarded if no message reorder occurs. Second, when q is reset, the number of discarded fresh messages is bounded by 2K_q. In either case, no replayed message will be accepted by q.

With the rapid growth of the Internet into a global communication and commercial infrastructure, the need for Quality of Services (QoS) in the Internet becomes more and more important. With a Bandwidth Broker (BB) support in each administrative domain Differentiated Services (DiffServ) is seen as a key technology for achieving QoS guarantees in a scalable, efficient, and deployable manner in the Internet.

This paper presents the design and implementation of a new Bandwidth Broker (BB) model to achieve QoS across a DiffServ domain. Our BB uses centralized network state maintenance and pipe-based intra-domain resource management schemes. The proposed model significantly reduces admission control time and minimizes scalability problems present in prior research while optimizing network resource utilization. The experimental results verify the achievements of our model.