[DO NOT MERGE] bench: block v2 iterator bench

[MrCroxx]

Key format: | 't' | tid (u32) | pk (u64) |
Value format: v (u64)
Each block contains 10 tid, each tid contains 100 entries (in order).
BlockV2 restart point interval: 16

Under the circumstances:

• BlockV2 encoding needs only half of the space of Block.
• But the performance of iterators regressed. But the performance of prev regressed.
• The regression of next may because copies of my implementation.
• The regression of prev, in theory, is next * (restart point interval / 2)

UPDATES: After optimization, BlockIteratorV2::next() is better than the origin one.

[MrCroxx]

[MrCroxx]

Seems the regression comes from Bytes::slice(), after removing unnecessary Bytes::slice(), BlockIteratorV2::next() shows better performance than BlockIterator::next().

commits:

You can download the svg and view them interactively.

b5964a8 (after optimization)

a32751f (before optimization, next only)

[MrCroxx]

Enough for now. Reopen anytime if needed.

[MrCroxx]

Close.