A model of service differentiation burst assembling and padding for improving transmission efficiency in OBS networks

Abstract

Service differentiation is an indispensable requirement for transmission in optical burst switching (OBS) networks, which can be based on offset-time, burst-length, or both, offset-time and burst-length. The offset time based approach sets a large offset time for high priority bursts and a small offset time for low priority bursts. Whereas, with burst length based approach, high priority bursts are short in size and low priority bursts are long in length. A combination of these two approaches promises to provide flexible service differentiation. The paper proposes a model of service differentiation burst assembling and padding, in which the assembly time threshold is set to reduce the end-to-end delay, but does not effect the burst length prediction accuracy; the length of generated bursts are flexibly adjusted based on feedbacked void size; and the high-priority burst with short length is padded by the lower priority data. Simulation results and analysis show that the proposed model is more efficient than previous similar models in terms of bandwidth utilization, byte loss, throughput fairness and estimation error.

A Hybrid Burst Assembly Algorithm Based on Transition Count Number for OBS Network

“…This method assembles a burst upon reaching either predefined Length or Time, whichever occurs sooner (13) . Extended-Timer-Based assembly algorithm and Fixed-Time-Min-Length assembly algorithm are the efficient versions of Time/Length hybrid approach (14,15) . There are some approaches in the literature which uses the incoming traffic rate as a parameter to set the sizeburst (16,17) .…”

“…Figure 4 shows the imposed traffic with H value to evaluate the performance of network. The performance of OBS model has been evaluated according to Time based threshold (Time) (26) , Length based threshold (Packet) (15) , Time/Length based threshold (E-Hybrid) (27) and traffic prediction based (Hybrid) (17) burst assembly as shown in Figures 5, 6 Every burst assembly strategy has two main objectives: the first is to reduce the packet delay during queue assembly in order to reduce the overall delay, and the second is to increase burst size. Large burst sizes do, however, result in fewer bursts produced, which minimises the processing load on the core nodes.…”