db_bench: many default values are not the default in Speedb

[Yuval-Ariel]

these params are:

table_cache_numshardbits is 4 while default is 6.
new table reader for comp inputs
block cache 8mb
index shortening mode
enable_pipelined_write - true in db bench and false otherwise.
delayed_write_rate = 8Mb and default value is 0.

and possibly more.

fix by making db_bench take the default value

[Yuval-Ariel]

changes:
index_shortening_mode - from 2 (kShortenSeparatorsAndSuccessor) to 1 (kShortenSeparators)
hard_pending_compaction_bytes_limit - from 128 Gb to 256 Gb
delayed_write_rate - from 8Mb to 0 (0 is sanitized to 16Mb)
enable_pipelined_write - from true to false

[erez-speedb]

Seems the rocksdb default are better then the db_bench defaults
Small obj
https://admin.speedb.io/performance?items=2R2iIbfzE1yWcnSIkwUQ&items=0YnC1biInuhalItBnuX6&colors=%23F06292&colors=%2300796b
Large obj
https://admin.speedb.io/performance?items=PDScQQlgGl9TOZfbs12T&items=DQbr0ntUjYJDz3RLExzO&colors=%23F06292&colors=%2300796b
Small DB
https://admin.speedb.io/performance?items=GFN4BI9Xr0wuX0MDjMF6&items=svLLRKKkMmxENz01Igpz&colors=%23F06292&colors=%2300897b
In small DB the overwrite of current default is 16% better but is all other configurations its the same. I belive is an outlier.

[Yuval-Ariel]

@erez-speedb , what do you mean outlier? the trend looks completely different

can you point to previous outliers that behaved like this? if you run two exact hashes, can there be a chance for such a difference?
why not run it again and verify whether its an outlier or not?

[erez-speedb]

Running the small db test again
The results are back to normal
https://admin.speedb.io/performance?items=MTnDOa9oyhGa33CyQgci&items=Swqg6xZSgqnVYc71L1mA&colors=%2300796b&colors=%23ec407a

Passed perf tests.