The main objective of this dissertation work is to investigate the performance of an ATM switch in broadband personalized multimedia networks.
Recently, we are moving into an age of personalized information networking. In this information age, communication is playing a more important role than ever before. Thus, we have to design a new communication network that has the integrated form of wired and wireless network based on broadband integrated service digital networks (B-ISDNs). In this work, we study the network architecture based on the ATM switch and the network control capability to allow creation of more intelligent services as well as existing broadband multimedia services. In addition, we study intensively performance improvement techniques of an ATM packet switch to provide advanced services such as broadband, multimedia, and multiparty group communication services, since it is the most important in network construction.
First, we propose the infrastructure of a future network and its control mechanisms. The proposed network architecture has a hierarchical structure that consists of access and transport network separately. The access network is the interface between the customers and the transport network. The transport network is implemented with standard B-ISDN networks composed of a high-speed ATM switching system and optical transmission systems. In addition, we propose a fully distributed network control of location registration and connection control based on a hierarchical addressing structure. This distributed mechanism can provide the advantages of network management and operation.
Second, we consider the window scheme which can improve the throughput of an input buffered ATM switch. We analyze this scheme in a single plane switch and obtain closed-form solutions of the maximum throughput. The analysis is done by the approximation technique of an equivalent queueing model. By analyzing the dynamics of HOL packets, we obtain the maximum throughput of the window scheme. Based on this result, we can also analyze the window scheme in a parallel-plane switch. The window scheme in a parallel-plane switch can improve the maximum throughput much higher. Thus, if we use a properly combined scheme of windowing and parallel-plane switching, it is possible to improve the maximum throughput with a small extra hardware cost.
Next, we consider the delay priority scheme in an ATM packet switch, which can improve the performance of delay sensitive traffics. Particularly, we study the priority effects on ATM packet switches with input buffer, and show various HOL priority schemes. Among them, we analyze some representative priority schemes, that is, nonpreemptive and preemptive schemes, and compare their performances with respect to the system throughput and cell transfer delay. For the nonpreemptive scheme, we approximate each HOL queue as a discrete-time M/D/1 queueing system with two priority classes employing random order of service discipline. For the preemptive scheme, on the other hand, we propose a new HOL queueing model with two HOL queues for low and high priority packets, separately. By analyzing these queueing systems, we can find that the preemptive priority scheme has more improved performance characteristic in a switching system than that of the nonpreemptive scheme for both low- and high-priority packets.
Finally, we propose a new cascaded type multicast switch, which consists of two independent subnetworks in tandem; space-division routing subnetwork and shared-medium multicast subnetwork. The major design consideration of this switch is that the copy network is succeeded by the routing network. By separating multicast and unicast traffics, the internal bandwidth requirement of multicast ring subnetwork is considerably reduced comparing to the shared-medium type multicast switch. However, it still has the overhead of speed-up to implement a large-scale multicast switch. Therefore, we also propose large-scale ATM multicast switches succeeded by a multicast copy network. This switch consists mainly of a newly proposed multicast cell distributing and copying network (MCDCN) and a set of multicast output switch modules. The MCDCN is composed of a sorting network and parallel dilated banyan networks based on the bit map addressing scheme. The key to design this network is the distributed control, self-routing and replicating capabilities. And, the use of a multicast switch with output buffer ensures the entire multicast switch to achieve high performance in terms of cell loss probability, delay and throughput. We approximately analyze the cell loss probability of this MCDCN, and show that the degree of dilation needs not be high to obtain the cell loss probability of $10^{-10}$.
최근 음성, 데이타 외에 영상과 같은 광대역 서비스와 이들이 서로 유기적으로 결합된 멀티미디어 그룹 통신 서비스의 시대를 맞이하고 있다. 뿐만 아니라 이동 통신을 근간으로 하는 개인 이동 통신 서비스와 이들이 결합된 지능망 서비스들이 새로이 요구 되고 있다. 본 논문에서는 이러한 요구에 맞게 다양한 서비스의 수용이 쉬운 유무선 결합 통신망의 연구를 통하여 광대역 멀티미디어 서비스와 함께 개인 이동 통신 서비스의 유기적인 결합을 쉽게 제공하기 위한 ATM (asynchronous transfer mode) 교환소자를 연구하고 성능을 분석하였다.
우선 영상과 같은 광대역 서비스 제공을 위한 ATM switch의 성능 향상 방법들에 관하여 연구하였다. 특히 본연구는 ATM switch 중에 구현이 쉽고 대용량화가 쉬운 input queueing switch의 성능을 향상시키는 한가지 방법으로 windowing기법에 관해 연구하였다. 또한 windowing 기법을 병렬형 switch에 적용하여 분석하였다. 이를 근간으로 병렬형 switch와 windowing 기법의 결합에 따른 성능 향상 정도를 비교 분석하였다.
다음으로 멀티미디어 트래픽과 같은 다양한 특성의 트래픽을 동일한 switch에서 제공할 때 switch 내에서의 트래픽 제어 기법을 통한 성능 향상 방법으로 time priority에 관하여 연구하였다. Switch에서의 priority 특성은 일반적인 queue 보다 훨씬 복잡하며 종류도 다양하다. 이들은 구현시 복잡성이나 성능 등에 차이가 있다. 본 연구에서는 이들간의 차이를 새로이 비교 분석하였다. 그 결과 switch에서의 priority는 기존의 queue 모델에서와 많은 차이를 발견하였다. 특히 preemptive 방법에서는 HOL에서의 work conservation law가 성립하지 않고 service time이 preemption된 low-priority 패킷과 preemption한 high-priority 패킷 간에 서로 겹치는 현상을 알아냈다. 뿐만 아니라 출력 경합시에 재시도에 따른 correlation 효과가 preemption에 의해 줄어드는 현상도 알아냈다. 따라서 preemptive priority 방법이 nonpreemptive 방법보다 high-priority 뿐만 아니라 low-priority 트래픽도 향상됨을 알았다.
끝으로, group 통신 서비스의 효율적인 제공을 위한 multicast switch 구조를 제안하고 성능을 분석하였다. 제안한 방법은 cascade ring multicast switch로 기존의 multicast switch와는 다르게 ring 형태의 copy network를 point-to-point 서비스을 위한 routing network 뒤에 붙임으로서 unicast traffic과 multicast traffic을 분리하여 copy network의 hardware complexity를 줄이면서 부가적인 packet 복제와 routing기능을 제공할 수 있는 구조이다. 또한 대용량의 switch 구성을 위한 구조를 새로이 제안하였다. 이 방법은 기존의 growable switch의 구조를 갖고 dilated banyan network을 통하여 self-routing과 packet 복제 기능을 제공한다. 따라서 전체적인 switch의 크기를 작은 module switch들로 분리할 수 있는 구조를 갖는다. 또한 module switch의 성능이 전체 switch의 성능을 나타내므로 output queueing 방식의 module switch을 통하여 전체 switch의 성능을 향상 시킬 수 있는 구조로 되어있다.
