H. Kariniemi and J. Nurmi (Finland)
ATM switch, RR, LQFS, RR-LQFS, arbiter, performance simulations
In internally buffered crossbar switches the total amount of the required buffer space is affected by the operation of the column arbiters that control the emptying of the cross point buffers. This paper compares the performances of three arbiters which are Round Robin (RR), Longest Queue First Served (LQFS) and RR-LQFS. The LQFS and RR-LQFS arbiter, that is a combination of RR and LQFS arbiters, can adapt its operation to the filling of the cross-point buffers. The performances are studied by comparing the distributions of the filling of the cross point buffers produced by the arbiters under different burst-traffic conditions. The spread of the distributions determines the overall buffer space requirements as well as the throughput of the switch. Average delay and delay variance that cells experience in the cross-point buffers are two other factors of the performance that are also studied in this paper. Simulation results show that LQFS and RR-LQFS generate almost equally wide distributions with both symmetrically and asymmetrically distributed loads enabling more area optimal implementation than RR which produces the widest distributions in every case. Simulation results show also that the RR-LQFS arbiter can partially solve the drawback of the RR and LQFS arbiters that they produce quite high buffering delays under certain asymmetric load conditions. The simulation arrangements have also been explained carefully in this paper including two different traffic generation models used in simulations. The effect of doubling the emptying rate of the cross-point buffers was also evaluated.
Important Links:
Go Back
