Packet Classification with Evolvable Hardware Hash Functions – An Intrinsic Approach

Rapid Evolution of Time-Efficient Packet Classifiers

“…Whether or not the number of conflicts vanishes depends on both the arguments and operators that can be employed into the hash function. Figure 3 sketches the evolvable hardware platform that has been developed in previous research [16]. The hardware works as follows: A key parser extracts the key, i.e., the destination IP address, from an incoming packet, and by means of a switch, forwards it to the hash function that is entirely realized in hardware.…”

“…3. This hardware classifier platform has been developed in previous research [16]. A genome feeds lg n elements, which select two bit positions from k-bit-wide keys by a number of multiplexers.…”

“…Thus, the optimization goal for the application at hand is to find the best one in a search space consisting of 2 2 lg(k) lg(n) =2 100 . For further implementation details, the interested reader 1 is referred to the literature [16].…”

“…Unfortunately, the routing profile of a single router changes over time, which would require adaptive hash tables in order to operate efficiently in time. Accordingly, previous research [16] provides a proof-of-concept that hash tables can be directly employed in hardware and that the required online adaptation can be achieved by an evolutionary algorithm, which can also be directly employed in the very same hardware. Therefore, this intrinsic (evolvable) hardware approach yields both mostly constant search time O(1) and constant online adaptation to changing routing profiles.…”

“…With n denoting the [16][17][18][19][20][21]2006 number of rules in the data base, a sequential search through the data base requires an average of n/2 comparisons resulting in a search complexity of O(n). With a steady increase in both the bandwidth and the number of ports, the size of the data base and thus the number of data base lookups are also steadily increasing.…”