config.py: run dataset-serialize on ursa-i9-9960x

[jgehrcke]

I have added a new benchmark type which focuses on Arrow Table serialization (to csv, arrow, parquet, ...) and writing to tmpfs:

• https://github.com/voltrondata-labs/benchmarks/blob/16330feaeaa9d498f0fa08794045270798e118c1/benchmarks/dataset_serialize_benchmark.py
• benchmarks: add dataset_serialize_benchmark.py benchmarks#124

Let's run this on ursa-i9-9960x.

[jgehrcke]

Landing this, let's see how this goes.

[jgehrcke]

This ran on agent ursa-i9-9960x.

Here's an example report fragment:

    "stats": {
        "data": [
            "29.978293",
            "30.135311",
            "30.120355",
            "30.122166",
            "30.128369",
            "30.132353"
        ],
        "iqr": "0.010550",
        "iterations": 6,
        "max": "30.135311",
        "mean": "30.102808",
        "median": "30.125268",
        "min": "29.978293",
        "q1": "30.120808",
        "q3": "30.131357",
        "stdev": "0.061269",
        "time_unit": "s",
        "times": [],
        "unit": "s"
    },
    "tags": {
        "cpu_count": null,
        "dataset": "nyctaxi_multi_parquet_s3",
        "format": "parquet",
        "name": "dataset-serialize",
        "selectivity": "100pc"
    },
    "timestamp": "2022-12-20T10:57:59.326723+00:00"

For all the reports I looked at, I saw:

• lovely stability across iterations.
• the first sample was the fastest one, slightly (so that's an interesting effect worth understanding).

All case permutations got written to cobnench.ursa.dev, see https://conbench.ursa.dev/runs/e471e650ae434930b3a2366ebf0b8bcc/