Performance and evaluation of adaptive backoff schemes in traffic shaping over high speed network

Abstract

Traffic shaping mechanism wants smooth traffic but network congestion built up in outgoing at high speed network still occurs when source traffic always becomes at hit the highest point rate. It is causing a lot of non-conforming frames. However, as one way to solving this problem, we purpose a backoff time in leaky bucket over traffic shaping mechanism. Backoff time computations are wildly used in order to avoid the impact on the network performance whenever the increment of offered load arises. The fluctuation of offered load will cause the congestion at waiting room. Furthermore it will lead to or will be close to the deadlock situation as many attempts are there for all-the-time busy service center. The backoff computation will help reduce the repetition of I/O cycles (required to switch task between waiting room to service center) and will prevent the maximum attempts to be reached due to the congested traffic at the waiting room, backoff time computation schemes proposed in this paper are pseudorandom backoff (PB) time, exponential backoff (EB) time and random backoff (RB) time. In pseudorandom backoff (PB) scheme, none of the computation is applicable but the queue discipline. In this paper, the FIFO and the maximum queue size are preset. In exponential backoff (EB) scheme, each node will double the backoff time interval up to the maximum backoff time after each attempt. On the other hand, it will decrease the backoff interval time to the minimum value after a successful attempt. In random backoff (RB) scheme, the delay in queue will be computed randomly, meaning that all frames are to wait in queue until next attempt arises for any random period of time. The performance of three backoff schemes has been investigated by the fluctuation of telecommunication traffic stream (ON/OFF stream). Simulation results indicate that EB and PB help improve performance of network depending on maximum waiting time in queue. Moreover, RB is found to be better regardless to QoS and better if required bandwidth are substantial.