Graduate Student: X. Zhang
Faculty: Kang G. Shin
Sponsor: ONR
One of most important issues in designing an ATM network is flow control which is essential for performance-guaranteed services and also flexible & efficient use of network resources. The propagation delay in high-speed channels of the ATM network limits the applicability of many existing flow control schemes, and the simple high-speed protocols used in ATM networks make it difficult to implement efficient and flexible flow control schemes. Among the many proposed, rate-based and credit-based flow control schemes are receiving significant attention. Using the rate-based flow control scheme, the network's end-to-end delay can be bounded. However, buffer overflows may occur if some users generate more traffic than specified at the connection setup time. In addition, the link bandwidth utilization cannot be maximized for every individual link along the virtual circuit. On the other hand, in the credit-based scheme, flow control is done in a link-by-link per virtual circuit fashion. Using the credit-line scheme, each node cannot forward traffic more than the next downstream node is able to accept. Consequently, buffer overflow and hence cell losses can be totally avoided. Moreover, by the "credit-updating" flow control scheme, one node can send early or extra cells whenever the buffer space for these cells are available at a downstream node. Thus, the bandwidth and buffer utilization per link can be maximized. Unfortunately, the credit-based flow control lacks the capability of bounding network end-to-end delay. Therefore, neither of the two schemes can fully satisfy the service requirements of ATM networks. We are working on a new flow control scheme which combines the merits and overcomes the disadvantages of the above two schemes. The features of this new flow control scheme includes: (1) maximizing link bandwidth utilization; (2) bounding end-to-end delays; (3) totally avoiding buffer overflow; (4) minimizing the buffer requirements.