We have an opportunity to rethink, from scratch, the design of our data structures. New byte-addressable non-volatile memory (BNVM) technologies promise the construction of systems with large persistent memories, potentially improving reliability and performance. With these technologies come new characteristics that deviate from those of flash and spinning disk—and with new characteristics come new optimization goals. In particular, the read/write cost disparity and fine granularity of updates allows us to save power and wear by reducing the bits flipped during writes to memory. Targeting these optimizations by formulating new data structure design and implementation strategies instead of relying on existing ideas will be vital for BNVM technology to reach its full potential. We modified a full-system simulator to count bit flips during program operation, opening the door for future research to design, construct, and evaluate data structures for these new goals.
