A flow distribution technique associates probabilistic data structure with each processor of a multi-processor system that is used to determine whether a received packet is included in a flow assigned to the associated processor. A set of tuples of a received packet is processed by multiple hash functions to generate a set of values utilized as indices to search for hits of the probabilistic data structures. Only those flows having hashed indices that hit more than one probabilistic data structure are saved in a collision flow table. The probabilistic data structure require much less memory space than a flow table and can be designed so that the frequency of collisions is very low thereby requiring a small amount of memory for the collision flow table.