The sliding-window approach to high performance ATM switching for broadband networks

Abstract

Typical design of a large shared buffer-based ATM switching system has been severely restricted by the bottleneck created by high memory bandwidth requirements, segregation of the buffer space and degradation of its throughput performance as the switch grows in size. In order to preserve its ability to provide for minimum possible cell-loss rate for a given buffer size, a shared buffer based ATM switching network design should attempt to provide for global buffer sharing among all its inputs and outputs. However, the complete sharing of a finite buffer space does not necessarily guarantee the best possible throughput performance. It is important for a shared buffer switch to use some kind of buffer sharing scheme to achieve the best possible performance. We present the sliding-window ATM switch architecture, its underlying switching scheme and discuss its memory requirements. The sliding-window switching scheme facilitates high throughput and an efficient sharing of a finite buffer space among all the switch inputs and outputs. Because of its efficient memory management, the sliding-window switching scheme can provide complete sharing, partial sharing and also complete partioning of a finite buffer space among its inputs and outputs.