Distributed-Queue Dual-Bus (DQDB) has been adopted by the IEEE 802.6 Committee as a medium access control protocol for high speed Metropolitan Area Networks (MANs). The main advantages of DQDB are the simplicity of its medium access control and that it can utilize all of the channel bandwidth independent of data transmission rate and network size. However, DQDB shows an unfair behavior in bandwidth sharing. Various fairness control methods have been studied to resolve this unfairness. The network also has a throughput limiting feature in which busy slots must continue to propagate downstream, even if they have already been read. This limitation has been resolved by introducing a slot erasure function that recognizes packets that have passed their destination stations, and releases the slots for subsequent use. However, this DQDB with slot reuse (DQDB-SR) also suffers from a location-dependent unfairness in access network and throughput deterioration due to the low efficiency of slot reuse, especially under heavy load conditions.
This paper studies the behavior of DQDB-SR and proposes two methods to improve the performance of DQDB-SR. The unfairness and throughput deterioration phenomena in DQDB-SR are, in first, investigated. DQDB-SR is generally implemented by introducing erasure nodes that divide the network into multiple segments. When the offered load exceeds the maximum throughput level of DQDB-SR, slot requests from downstream segments begin to flow toward their upstream segments. This request overflow incurs a segment unfairness that forces stations on the upstream segments to be starved in access to the network. The efficiency of slot reuse will then decline, which leads to a throughput deterioration.
The first method to improve the performance of DQDB-SR is to use bandwidth access control techniques. The bandwidth balancing mechanism (BWB) and the access protection scheme (APS) used in DQDB are considered. To apply the APS, the optimal placement of erasure nodes and an upper protection limit needed to guarantee each station''s bandwidth sharing are determined. A proper bandwidth balancing modulus for BWB is obtained through simulation. Their performances are compared under various workload types. Simulation results show that the APS is quite suitable for the bandwidth access control to improve the performance of DQDB-SR at overload conditions.
The location-dependent unfairness of DQDB-SR is inevitable due to the nature of unidirectional bus architecture. In addition, since any slot can not traverse from the end of the bus to the head of the bus even though they are located closely together, throughput level is restricted compared with a ring network. A loop architecture of DQDB-SR (LDQDB-SR) has been studied to overcome those performance limitations. The LDQDB-SR adopts the destination slot release and an inter-segment bandwidth regulation based on the distributed queuing system of DQDB. The inter-segment bandwidth regulation is to distribute the network bandwidth to all segments. This network suffers not only from a severe throughput deterioration due to the high regulation cost and excessive transit delay but also from an unfairness in bandwidth sharing, especially under an overload condition. We introduce the quota-based fairness control used in a ring network with spatial reuse to enhance the LDQDB-SR. This is the second method to improve the performance of DQDB-SR. This network with the fairness control feature of a ring network is called ELDQDB-SR.
The fairness control of a ring network is to fairly distribute the network bandwidth to all nodes, and each node uses a packet window to restrict its packet transmission. Under this algorithm, satisfied nodes, which have exhausted their packet windows, stop their packet transmissions to pass empty slots to starved downstream nodes until their windows are refilled. Due to the nature of this regulation, empty slots arriving at each satisfied node can not be occupied by the node. Those empty slots are called the blocking bandwidth waste. A cycle window is introduced and a new fairness control scheme based on a two-layer window composed of the cycle and packet windows is proposed, under which the wasted bandwidth can be reused to carry, in advance, packets allocated to future fairness cycles. For an ideal ring, an analytic approximation method for the bandwidth utilization is developed. The new method is applied to the two typical ring protocols with only the packet window such as the ATMR and MetaRing. Simulation results show that all networks have their optimum performance at the cycle window of size 2.
For an inter-segment bandwidth control, the ELDQDB-SR uses the fairness control of MetaRing with the cycle window of size 2 because it shows the best performance. The inter-segment bandwidth control function is implemented at each erasure node. Each station selects the bus that minimizes the number of segments on the path to destination stations. The network capacity of ELDQDB-SR is analyzed under a homogeneous traffic load. To achieve a fair bandwidth distribution among stations within each segment, the alpha-tuning mechanism is generalized. Simulation results show that the ELDQDB-SR gives an enhanced throughput level and also maintains good fairness under overload conditions. This second method shows a superior performance than the first one, but it needs some hardware modification.