A. Small I/Os are aggregated in the cache and then written to disks, achieving an ideal latency and improving the utilization of small files.

B. Large I/Os can be directly forwarded to the home node for processing, eliminating the overhead of scale-out locks and greatly reducing the read/write latency of large I/Os.

C. Fixed-length large-granularity indexes are used to ensure sequential read/write performance of large I/Os.

D. In terms of I/O flow, large I/Os are written to disks in passthrough mode, improving the bandwidth capability.