benchmarks.json: tweak dataset-serialize invocation

[jgehrcke]

Changes:

• more iterations to get more conclusive statistics
• no need to drop filesystem / page cache because disk does not have involvement

I hope that this passes CI (i.e., I hope that the automatic check does not require --iterations=3 --drop-caches=true.

[jgehrcke]

E         -     "command": "dataset-serialize ALL --iterations=6 --all=true",
E         ?                                                    ^
E         +     "command": "dataset-serialize ALL --iterations=3 --all=true --drop-caches=true",
E         ?                                                    ^           +++++++++++++++++++
E               "flags": {
E                 "language": "Pyth

I find this rather inflexible, I think depending on the benchmark specifics we should be able to tweak things. What do we think?

I suspect that I can actually tweak these two parameters, elsewhere in code -- looking.

[jgehrcke]

Seems like iterations can be set as class attribute, and it seems like --drop-caches=true is hard-coded: https://github.com/voltrondata-labs/benchmarks/blob/16330feaeaa9d498f0fa08794045270798e118c1/benchmarks/_benchmark.py#L338

[coveralls]

Pull Request Test Coverage Report for Build 3739321263

• 1 of 1 (100.0%) changed or added relevant line in 1 file are covered.
• No unchanged relevant lines lost coverage.
• Overall coverage increased (+0.007%) to 87.222%

---

Totals
Change from base Build 3706037481:	0.007%
Covered Lines:	1488
Relevant Lines:	1706

---

💛 - Coveralls