mirror of
https://github.com/scylladb/scylladb.git
synced 2026-04-28 12:17:02 +00:00
We know the correct boundaries now, so we can use that information to feed the default buffer size: if it is small enough (smaller than 128k), we can try to bring everything at once. For the default key sized 128 that we use in the index read perf: (smp == 1, partitions = 500000, concurrency == 1) Before: 423493.26 +- 811.03 partitions / sec (30 runs, 1 concurrent ops) After: 707311.86 +- 1865.47 partitions / sec (30 runs, 1 concurrent ops) For a gain of 67 %. Signed-off-by: Glauber Costa <glommer@cloudius-systems.com>