Optimal flow control window size design in high-speed networks

Claim 3

A high-speed communications network, comprising: A plurality of interconnected network devices said network devices including an originating server, a receiving server and zero or more intermediate servers; and means for implementing flow control between the network devices, said flow control utilizing an optimal window size; whereby the network power is maximized, approximated by an upper bound on the window size of 4(M+T.sub.P /S.sub.M)+max{0, (T.sub.P /2S.sub.M)-1}, where M is the number of servers in the network, T P is the round trip propagation delay between the originating server and the receiving server, and S M is the average delay at the slowest server.

Claim 4

A high-speed communications network, comprising: A plurality of interconnected network devices said network devices including an originating server, a receiving server and zero or more intermediate servers; and means for implementing flow control between the network devices, said flow control utilizing an optimal window size; whereby the network power is maximized, approximated by an upper bound on the window size of 4(M+T.sub.P /S.sub.M)+max{0, (T.sub.P /2S.sub.M)-1}, where M is the number of servers in the network, T P is the round trip propagation delay between the originating server and the receiving server, and S M is the average delay at the slowest server.

Claim 5

The system described in claim 2, wherein said optimal window size is also bounded by a lower bound, said lower bound being: ##EQU8## where s i is the packet service time at server i=1,2, . . . M.

Claim 6

A network node for performing network flow control in a network of interconnected servers, said network flow control comprising the steps of: determining the number of servers in the network; calculating a round-trip propagation delay between an originating server and a receiving server; determining the average delay for a packet at the slowest of the servers in the network; calculating the optimal window size using the number of servers, the round-trip propagation delay and the average delay for a packet at the slowest server; and transmitting packets of data using the optimal window size to other servers in the network, wherein: the calculated optimal window size is 4(M+T.sub.P /S.sub.M)+max{0, (T.sub.P /2S.sub.M)-1} packets; where M is the number of servers in the network, T P is the round trip propagation delay between the originating server and the receiving server, and S M is the average delay at the slowest server.

Claim 7

A network node for performing network flow control in a network of interconnected servers, said network flow control comprising the steps of: determining the number of servers in the network; calculating a round-trip propagation delay between an originating server and a receiving server; determining the average delay for a packet at the slowest of the servers in the network; calculating the optimal window size using the number of servers, the round-trip propagation delay and the average delay for a packet at the slowest server; and transmitting packets of data using the optimal window size to other servers in the network, wherein: the calculated optimal window size is 4(M+T.sub.P /S.sub.M)+max{0, (T.sub.P /2S.sub.M)-1} packets; where M is the number of servers in the network, T P is the round trip propagation delay between the originating server and the receiving server, and S M is the average delay at the slowest server.