The main objective of this dissertation work is to investigate the performances of bandwidth allocation and bandwidth enforcement strategies for traffic controls in fast packet-switched broadband ISDNs(called ATM networks).
In ATM networks, network congestion is controlled by limiting the traffic flow into the network so as not to reach the level causing unacceptable congestion. This is done by a priori bandwidth allocation for calls at call set-up and by the enforcement of the allocated bandwidth during the call holding time. But, before applying specific bandwidth allocation and enforcement strategies, the estimation of the performance of each traffic is needed in order to satisfy the grade-of-service for each traffic and to achieve the maximum efficiency of the network.
For this purpose, we first studied the deterministic bandwidth allocation strategies for the integration of wide-band and narrow-band traffics. As a traffic control strategy in the deterministic bandwidth allocation mode, we set the access restriction on each type of traffic to protect overloading of one traffic and meet the GOS for each type of traffic. Also, as an improved strategy, we proposed a strategy that allows overflowing of bandwidth allocations over the access restrictions to use the idle channels reserved for there other traffic. We analyzed these allocation strategies using the matrix-analytic solution approach, and obtained the system time for NB calls, the blocking probability for WB calls, and the power combining the performance of the NB and WB traffics. Numerical results showed that the performance depends on the ratio of mean service times, the ratio of bandwidth requirements, each traffic load, and the cutoff values. These strategies can adapt to the varying traffic loads easily to meet the GOS required for each type of traffic by simply changing the restriction levels. We also showed that these strategy may be regarded as generalizations of various extreme strategies.
Next, we analyzed the performances of the heterogeneous bursty traffics under the statistical bandwidth allocation from the bursty traffics. We modeled each bursty traffic with a Markov-modulated Poison process and represent the different burstiness of each traffic by changing the parameters of the MMPP. Then, we used a finite queue multi-server model for the statistical multiplexer with multiple output channels. In order to consider the slottized operation of the ATM channel, we also assumed that the transmissions of the cells queued in the buffer occur at discrete time instants. With these traffic and system models, we obtained the cell delay distribution and the cell blocking probability for the superposition of heterogeneous bursty traffics. We also measured the performance of each traffic separately with the cell delay and the cell blocking probability for each traffic. From numerical results, we showed the effects of each traffic's characteristic on the performance of the statistical multiplexer. We noted that the burstiness of the MMPP increases as the difference between arrival rates becomes large and as the state duration times become long. And the cell delay and the cell blocking probability increase as the burstiness of the traffic increases. In the case of integrating heterogeneous bursty traffics, the performance of each traffic is different for each other and it is worse as the burstiness of the traffic increases. From these results for the statistical bandwidth allocation strategy, it has been found that it is efficient to divide those traffics with comparable burstiness into segregated groups and allocate bandwidth to them exclusively.
Finally, we analyzed the performance of the bandwidth enforcement strategy applied for the bursty traffic. As a bandwidth enforcement strategy, we considered the leaky-bucket strategy which has been regarded as the frost promising and easily implementable strategy. In this analysis, we also used an MMPP as a model for the bursty input traffic. With a token and a token pool as models for the transmission right and the allowed burstiness of the input traffic, respectively, we analyzed and obtained the performance of the traffic under the LB bandwidth enforcement strategy. We also presented some numerical results to show the effects of the system parameters on the performance of the traffic measured by the cell delay and the cell blocking probability. From these numerical results, we could see that there is a trade-off between the cell delay and the cell blocking probability and it can be controlled by the input buffer for cells. We also found that the performance of the traffic is improved as the size of the token pool increases, but properly limiting the size of the token pool is necessary for enforcing the traffic not to cause the network congestion.
본 논문에서는 고속 패킷 교환 방식에 의한 광대역 종합정보통신망 (CCITT에서 ATM 망으로 명명됨)에서의 트래픽 제어를 위한 대역폭 할당 및 대역폭 이용 제어 방법들에 대하여 성능 분석을 수행하였다.
ATM망에서의 congestion 제어는 입력 트래픽이 congestion을 일으킬 수 있는 수준이 되지 않도록 각 호 개설시 선행적으로 적당한 대역폭을 할당하고 이를 초과하지 않도록 이용상황을 감시 제어하여 이루어진다. 그러나 특정한 대역폭 할당 및 이용제어 방법들을 사용하기 전 각 트래픽에 대한 성능예측이 각 트래픽이 요구하는 서비스 수준을 만족시키고 망전체 효율을 극대화시키기 위하여 필요하다.
이러한 필요성에 따라 본 논문에서는 첫째로 광대역과 협대역 트래픽을 결정적 대역폭 할당방식 (detrerministic bandwidth allocation mode) 하에 집적시킨 시스템에 대하여 연구하였다. 결정적 대역폭 할당방식하의 트래픽 제어방법으로 각 트래픽의 사용량을 제한하여 특정 트래픽의 과부하를 방지하고 각 트래픽의 서비스 수준을 만족시켰다. 또한 보다 개선된 방법으로서 타 트래픽이 사용하고 있지 않는 채널이 있을 경우 channel 사용을 제한량 이상이라도 허용하는 방법을 제안하였다. 이들 대역폭 할당 방법들을 matrix-geometric 방법으로 분석하여 협대역 호의 시스템 시간과 광대역 호의 blocking 확률 및 이들 두 값을 종합한 power등을 구하였다.
다음으로 이질의 bursty한 트래픽이 통계적 대역폭 할당방식 (statistical bandwidth allocation mode) 하에 집적된 시스템의 성능을 분석하였다. 각 트래픽은 Markov-modulated Poisson process로 모델링하였고 다중 출력 채널의 통계적 다중화 시스템은 finite queue multi-server 모델을 사용하였다. 또한 슬롯화된 ATM 출력 채널을 위해 각 cell들의 전송은 discrete한 시점에서만 이루어진다는 가정을 하였다. 이러한 트래픽 및 시스템 모델 하에 이종의 bursty한 트래픽이 집적된 입력으로 부터 발생한 cell들이 격는 지연시간 및 blocking 확률을 구하였다. 또한 서로 다른 각 트래픽이 격는 성능을 각 트래픽별로 구하였다.
마지막으로 대역폭 이용 제어하의 bursty 트래픽의 성능을 분석하였다. 대역폭 이용 제어 방법으로는 현재까지 ATM망에 가장 적합한 것으로 알려진 leaky-bucket 방식을 택하였다. Bursty 트래픽의 모델에는 MMPP를 사용하였고 leaky-bucket 방식에서의 cell 전송 권한과 트래픽에 허용된 burstiness를 token 및 token pool로 각각 모델링하였다. 이러한 시스템의 성능을 cell 지연시간 및 cell blocking 확률로 측정하였으며 각 시스템 변수들이 이들 값에 미치는 영향등을 분석하였다.
