BloomyCAN: Probabilistic Data Structures for Software-defined Controller Area Networks

Software-defined Networking (SDN) is an increasingly relevant topic for modern automotive EIE-architectures. Despite Ethernet and IP, a large part of a vehicle's network comprises of low-power field-buses such as controller area network (CAN). SDN4CAN brought SDN to the CAN world by introducing a configurable element, the forwarding device, to CAN nodes. The forwarding element can store CAN arbitration IDs within white or blacklists to manage bus access. Storing large lists requires a lot of space, which is an important cost factor in CAN systems. An approach towards improved memory efficiency is to use probabilistic data structures like Bloom Filters. However, they bring their own challenges like presence of false positives, i.e., providing access to the bus for non-eligible CAN IDs. In this paper, we extend the SDN4CAN architecture to introduce BloomyCAN - a memory efficient extension of SDN4CAN based on probabilistic data structures, that exploits the presence of an all-knowing SDN controller to remediate the false positive challenges of the Bloom Filters.