An efficient buffer management policy based on an integrated Fuzzy-GA approach

This methodology approach allows the fuzzy system parameters to be automatically derived when the switch parameters vary, offering a high degree of scalability to the fuzzy control system. Its performance is very close to that of an ideal mechanism like the push-out mechanism, and at any rate much better than that of the threshold schemes based on conventional logic. In addition it is simple to implement and above all inexpensive when implemented using VLSI technology. A. INTRODUCTION One of the most significant problems in ATM packet-switched architectures is buffer management. The various policies investigated in the literature include input queuing, output queuing and shared buffering [1][2]; the best performance in terms of delay-throughput is obtained by schemes based on output queuing with completely shared buffering: each output port has its own logical queue but shares the same memory area with the other ports. There are still, however, a number of open issues to be addressed in these schemes when specific buffer management policies are not adopted. One of these is performance degradation due to hot spot load conditions on an output port. In these conditions, the port takes up most of the memory, with a consequent loss of packets headed to other, less heavily loaded, ports and a reduction in the total throughput of the switch. To tackle this problem, various buffer management schemes operating on a shared memory have been proposed [3][4][5][6]. The schemes that various studies have shown to provide the best performance are those based on push-out policies. As long as there is room in the buffer, push-out policies always accept new cells that arrive; otherwise, cells already in the buffer are selectively replaced with new arrivals. We do not question the rules a policy has to follow in choosing the cells to be pushed out, like drop on demand (DOD) [3], push-out with threshold (POT) [4] or maximum busy period algorithm (MBP) [6], and we accept the fact that a push-out policy generally has a lower packet loss ratio than other policies which do not allow cells already in the buffer to be discarded (non-push-out policies). In this paper we consider a version, simply called Push-Out (PO), which, if the shared memory is full on the arrival of a packet, scans the various output queues in the shared buffer and pushes out the packet at the head of the longest queue. The advantage of this scheme …