Control of packet flow in statistical data forks

This paper presents a new structural property of the statistical data fork (SDF) which is exploited to derive a method for controlling packet flow in SDF. The property states that among all output channels, the one labeled with a binary number i can be made the most likely recipient of the packets associated with the codeword cⁱ which is defined as the bit reversal pattern of i. The property also states that no other packets with a different codeword appear at this channel with a greater probability. A sufficient condition for this property to hold is for the probabilities at the input to be certain permutations of each other. Using this property, a convenient addressing scheme is presented for assigning channel groups in the SDF when the input probability is uniform. In the proposed addressing scheme, an output channel can be assigned to a channel group based only on its address bits and this assignment is optimal for all levels of the offered load to the SDF