The main objective of this dissertation work is to investigate the performance of traffic control schemes for the high-speed integrated broadband networks.
The ATM networks must accommodate diverse types of applications with different quality of services (QOS) and traffic characteristics, which maybe require point-to-point connections or multipoint connections. Thus, in order to increase the utilization of the networks and satisfy the QOS's of each traffic type, some types of traffic control schemes must be adopted in ATM networks. And the accurate estimation of the performance of traffic control schemes is needed for dimensioning of the networks, determination of control parameters, connection admission control, etc.
First, we study the performance of a nonblocking input and output queueing multicast ATM switch with capacity C. A multicast packet switch provides an efficient transport method which can connect any input port to any subset of output ports. Each input packet of the switch is copied to multiple packets with identical payload and destined to a subset of output ports. So, the offered load to output ports increases as the mean number of copy packets generated by one input packet increases in a multicast switch, thus requiring some scheme of speeding up the switch. Our study indicates that the service time involved by packets at the head of input queues becomes a phase-type (PH) process for a large-size switch. With this service time distribution, we analyze the saturation throughput of the switch, input queue length distribution and mean delay using the matrix-geometric solution method. Our numerical and simulation results indicate that the switch capacity 2 is sufficient for adequate multicast switch performance. Also we study the performance of a multicast ATM switch with capacity C which adopts either one of two fanout splitting disciplines, i.e., the first-come-first-served (FCFS) head-of-line (HOL) service discipline and the random HOL service discipline. We analyze the service time distributions of a copy packet and an input packet. Using this distribution, the saturation throughput of the switch, mean delay and input queue length distribution are obtained. By numerical analysis and simulation, we show that the FCFS HOL service outperforms the random HOL service.
Second, we consider a nonblocking input and output queueing ATM switch with capacity C in which two classes of traffics with different loss probability constraints are admitted. Each input queue adopts either one of two space priority mechanisms, i.e., partial push-out and partial buffer sharing, which assign priority for buffer access to each traffic type according to the buffer occupancy status. We obtain the distribution of input queue length and loss probabilities of each traffic class using the matrix-geometric solution method and compare their performances. Numerical analysis and simulation indicate that the utilization of the switch with space priority mechanisms satisfying the QOS's of each traffic class is much higher than that of the switch without control and the required buffer size is reduced while satisfying the same QOS's. The partial push-out scheme has somewhat higher throughput than the partial buffer sharing scheme but the latter is preferred to the former due to its lower implementation complexity.
Next, we study the performance of a nonblocking input and output queueing ATM switch with capacity C which adopts a hybrid priority control of delay and loss priority. Each input queue of the switch has two separate buffer spaces with different sizes for two classes of traffics and adopts a general state-dependent scheduling scheme which assigns priorities for the order of service to each class dynamically. Especially, we consider 5 scheduling schemes and compare their performances. We obtain the input queue length distribution, loss probabilities and mean waiting times of two classes of traffics using a semi-Markov process concept. We show by numerical examples and simulations that one can control the performance of two classes of traffics flexibly and effectively using the above hybrid priority control scheme, and thus satisfy the QOS's of each traffic while achieving the maximum utilization of the switch.
Finally, we propose a new input rate regulation algorithm in which the credit generation rate is changed according to the states of an on-off data source and buffer occupancy. The leaky bucket (LB) algorithm which regulates the input traffic is a promising usage parameter control (UPC) algorithm for preventive congestion control. But, in order to satisfy the QOS of a bursty traffic, the LB algorithm requires a large data buffer and credit pool size. This in turn increases the reaction time of the LB algorithm to detect the violations of the traffic contracts and take corresponding actions, and the mean delay for an input traffic to pass the UPC device, which would be inappropriate for real time traffics. The proposed dynamic rate leaky bucket (DRLB) algorithm requires smaller buffers than the static-rate-algorithm to satisfy the same QOS's. It can also prevent the bandwidth from being wasted by the source which are too demanding. Using the uniform arrival and service (UAS) model, the performance of the proposed algorithm is analyzed.
다양한 서비스 품질(QOS)과 트래픽 특성을 가지는 여러 종류의 트래픽을 효과적으로 수용하기 위해 광대역 asynchronous transfer mode (ATM) 망에서는 여러가지 트래픽 제어방식을 채택하고 있다. 본 논문에서는 광대역 ATM망에서 트래픽 제어 방식들에 대해 정밀한 성능 분석을 행하고, 이를 이용해 그 효용성에 관해 논하고 제어 parameter들을 결정할 수 있음을 보였다. 먼저, 용량 (capacity) C를 가지는 multicast ATM 교환기에 대해 복사된 패킷을 임의로 선택해 전송하는 random HOL 서비스 방식의 성능을 matrixgeometric 해법에 의해 구하고, 적당한 multicast 교환기 성능을 위해서 용량 2인 교환기로도 충분함을 보였다. 또한 multicast 교환기에 대해 HOL 위치에 먼저 도착한 패킷을 먼저 전송해 주는 FCFS HOL 서비스 방식에 대해 성능분석을 하고 random HOL 서비스 방식의 성능과 비교했다. FCFS HOL 서비스 방식이 HOL blocking 효과의 감소에 의해 random HOL 서비스보다 더 좋은 성능을 보였다. 다음으로, 용량 C를 가지고 입·
출력단에 버퍼를 가지는 ATM 교환기에서 허용손실률이 서로 다른 두 트래픽에 대해 두 가지 손실우선순위 제어 (push-out 방식, 부분 버퍼 공유 방식)를 행할 때의 성능을 분석하고 각 방식의 장·단점을 논하였다. 우선순위 제어에 의해 교환기의 효율을 높일 수 있고 버퍼를 줄일 수 있음을 보였다. 같은 허용 손실률에 대해서는 전 버퍼영역을 공유할 수 있는 pushout 방식이 좀 더 높은 최대허용부하를 나타냈다. 하지만 이 방식은 버퍼내의 각 패킷의 위치를 모두 추적해야하는 단점을 가지므로, 버퍼의 임계값 (threshold)에 따라 유연한 성능을 가지는 부분 버퍼공유 방식을 채택하는 것이 더 합당하다고 볼 수 있다. 다음으로, 다양한 허용 손실률과 허용 지연시간을 가지는 트래픽을 제어하기 위해 손실 우선순위 제어와 지연 우선순위 제어를 동시에 행하는 복합 우선순위 제어 방식을 입·출력단에 버퍼를 가지는 ATM 교환기에 적용한 경우의 성능을 분석하고 교환기의 성능을 높일 수 있음을 보였다. 지연 시간 제어를 위해 생각한 다섯가지 상태의존 scheduling 방식중에 queue length threshold (QLT) 방식이 가장 뛰어난 성능을 보여 주었다. 마지막으로, 유망한 usage parameter control (UPC) 알고리즘중에 하나인 leaky bucket (LB) 알고리즘이 burst 트래픽에 대한 감시를 좀 더 효율적으로 할 수 있도록 하기위해 dynamic rate leaky bucket (DRLB) 알고리즘을 제안하고 성능을 분석했다. DRLB 알고리즘은 기존 LB 알고리즘보다 적은 버퍼와 토큰 pool로도 burst 트래픽에 대한 감시, 제어효과가 뛰어나고 지연 시간과 규약위반을 감지하는데 걸리는 시간이 짧아 졌다. 또한 과부하시에 낭비되는 bandwidth를 줄일 수 있었다.
