Yah this has been going on for a while. Before SPDK it was with custom kernel bypasses and fast inifiband/FC arrays. I was involved with a similar project in th

Yah this has been going on for a while. Before SPDK it was with custom kernel bypasses and fast inifiband/FC arrays. I was involved with a similar project in the early 2000's. Where at the time the bottleneck was the shared xeon bus, and then it moved to the PCIe bus with opterons/nehalem+. In our case we ended up spending a lot of time tuning the application to avoid cross socket communication as well since that could become a big deal (of course after careful card placement). But SPDK has a problem you don't have with bypasses and uio_ring, in that it needs the IOMMU enabled, and that can itself become a bottleneck. There are also issues for some applications that want to use interrupts rather than poll everything. Whats really nice about uio_ring is that it sort of standardizes a large part of what people were doing with bypasses.