Massively Parallel Identification of Privacy-Preserving Vehicle RFID Tags

Abstract

This article proposes a massively parallel identification scheme of vehicle RFID tags. These tags use a pseudo-random identifier, which is the output of a hash function fed by a fixed secret key that uniquely identifies the tag and by two random challenges that change on each tag activation. The use of random challenges makes it extremely difficult for someone not knowing the secret key of a tag to track its multiple activations. For someone knowing all valid keys, finding out the key that generated a specific tag response requires a time-consuming exhaustive search, if the number of valid keys is large. This can be performed in a very efficient way on a general purpose graphics processing unit. Our simulations show that on a very demanding scenario a single Tesla S1070 system can identify in near real-time the tags generated by 100 single-lane highway RFID readers.

A Code Skeletons for Kernel Termination After a Match

Fig. 3.

CUDA code skeletons for terminating a running kernel.

Figure 3 illustrates how to terminate a kernel on a single device (top) or on multiple devices (bottom) after a match; ... denotes omitted irrelevant code. On the latter case, our solution is only effective if the test is done inside a loop (in our case, this corresponds to each thread checking many keys). Given that accessing memory-mapped host memory from the device is very slow (the transaction is over the PCIe bus), on each round only one thread of each block does the access, and stores the value read in fast shared memory. To keep things balanced, the thread doing this slow access is changed as the computation progresses (an increment of $32$ corresponds to a jump to the next warp).