Reducing Energy Consumption of HMAC Applications on Heterogeneous Platforms

Hash-based message authentication code (HMAC) is a form of cryptographic checksum which is frequently used in SSL/TLS. The IETF standard permits several hash function families with several possible checksum lengths. Thus, if checksums must be computed for a number of messages, a multi-dimensional optimization problem arises, especially if a heterogeneous platform like ARM big.LITTLE is used. We formulate the optimization problem as an energy minimization problem with constraints given by deadlines and the different power and performance profiles of different hash functions on different core types. Furthermore, we show that with the help of measurements done in previous work this optimization problem can be mapped to a known scheduling problem for big.LITTLE, i.e., that a scheduling algorithm for that problem can be adapted to our scheduling problem. Thus, a solution can be computed that, given deadline as well as message sizes and hash functions for the HMAC computations, minimizes energy consumption on the big.LITTLE, and takes profit from the heterogeneity. We compare the scheduling results from an optimal and a heuristic scheduler and find that the heuristic on average produces schedules with 4% higher energy consumption, with small variance.