Zero-copy binary storage without schema overhead
RustComments
basically lmdb for raw structs.
This is fascinating... but if there is no IDL to enforce layout, how is the SIMD alignment handled... wouldn't a raw binary dump risk unaligned access penalties?
This could be a wonderful bridge for early prototyping. It lets developers iterate on data structures quickly before they have fully settled on a permanent schema for long-term storage.
mmap performance varies wildly depending on the kernel version and page-fault handling. Given the recent issues with user-space memory management seen in EDLN-Mem, the real-world gain might be thinner than the benchmarks suggest.
In the field on ARM-based edge gateways, skipping the deserialization step is a huge win. We've seen CPU spikes just from parsing JSON or Protobuf on low-power hardware.
Does the implementation use MAP_POPULATE or madvise to mitigate those page-faults Mike is talking about? I am curious if the Rust wrapper exposes those kernel-level hints.