Json input format

[da-ekchajzer]

{
"contributor": @da-ekchajzer,
"date": "date",
"write_mode": auto,
"hardware" : {
    "disks": [
        {
            "capacity": 489,
            "manufacturer": "Crucial_CT525MX3",
            "type": "ssd"
        }
    ],
    "cpus": [
        {
            "vendor": "GenuineIntel",
            "name": "Intel(R) Core(TM) i7-7500U CPU @ 2.70GHz",
            "microarch": [
                "kabylake",
            ],
            "core_units": 4,
            "family": "Kabylake"
        }
    ],
    "rams": [
        {
            "capacity": 7
        }
    ],
    "mother_board": {}
}

"content" : [
{
"timestamp": "date",
"power":100,
"load": 40,
"temperature":34,
"power_type": "RAPL",
"stress_type": "stres-ng x"
},
{
"timestamp": "date",
"power":230,
"load": 67,
"temperature":43,
"power_type": "RAPL",
"stress_type": "stres-ng x"
}
...
]

}

[maethor]

For consumption data, I suggest :

server_id, timestamp, rapl_power, dcmi_power, sensor_power, other_power (some power given by the user), load1, temperature (which one?), disks_io, ram_usage, ram_io?, cpu %user, cpu %sys, cpu %iowait, netstats?, and everything that can be easily measured and that can impact the power consumption.

[samuelrince]

JSON schema:

{
    "device_id": null,
    "contributor": null,
    "hardware": {
        "cpus": [
            {
                "name": null,
                "core_units": null
            }
        ],
        "rams": [
            {
                "vendor": null,
                "capacity": null
            }
        ],
        "disks": [
            {
                "type": null,
                "vendor": null,
                "capacity": null
            }
        ]
    },
    "states": [
        {
            "type": null,
            "powers": [
                {
                    "source": null,
                    "scope": null,
                    "value": null
                }
            ],
            "temperature": null,
            "disks_io": null,
            "ram_io": null,
            "cpu_percent_user": null,
            "cpu_percent_sys": null,
            "cpu_percent_iowait": null,
            "netstats": null
        }
    ],
    "date": null
}

[maethor]

I did not succeed to get ram_io, and I do not get netstat (yet), but I have a script that returns this.

{
  "interval_us": 5270833,
  "duration_us": 63249999,
  "nb_states": 12,
  "cpu_iowait_pct": 0,
  "cpu_sys_pct": 11,
  "cpu_usr_pct": 15,
  "load1": 4.44,
  "mem_free_MB": 20892,
  "mem_total_MB": 32065,
  "mem_used_MB": 5327,
  "sda_pct_busy": 1,
  "sda_read_bps": 2,
  "sda_write_bps": 18,
  "sdb_pct_busy": 1,
  "sdb_read_bps": 2,
  "sdb_write_bps": 17,
  "sdc_pct_busy": 1,
  "sdc_read_bps": 1,
  "sdc_write_bps": 19,
  "sensors_coretemp": 46,
  "powers": {
    "dcmi_cur_watt": 26,
    "rapl_dram_0_watt": 2,
    "rapl_package_0_watt": 14,
    "rapl_total_watt": 16,
    "sensors_acpi_watt": 25
  },
  "energizta_version": "0.1a"
}

interval_us is the interval in microseconds between each measure, and "duration_us" is the total duration of the test. Here I measure every 5 seconds, during 60 seconds, so the output is an average of 12 measures.

With this method I get very consistent results and I dont see "inexplicable" spikes in DCMI Current Watt.

I have simplified the "powers" dict because I believe the scope can be deducted from variable names.

PS : We can see each measure takes 270ms (on a 3.7Ghz server that's loaded to around 30 to 50%). I am not happy with this but it is open to improvement.

And this is the result of a stresstest with 4 steps (no stresstest, stress-ng -q --cpu 1, 4 and 8)

{"interval_us": 5381796,"duration_us": 43054361,"nb_states": 8,"cpu_iowait_pct": 0,"cpu_sys_pct": 12,"cpu_usr_pct": 16,"load1": 1.68,"mem_free_MB": 20857,"mem_total_MB": 32065,"mem_used_MB": 5358,"sda_pct_busy": 2,"sda_read_bps": 2,"sda_write_bps": 21,"sdb_pct_busy": 2,"sdb_read_bps": 2,"sdb_write_bps": 21,"sdc_pct_busy": 2,"sdc_read_bps": 1,"sdc_write_bps": 23,"sensors_coretemp": 42.0,"powers": {"dcmi_cur_watt": 21,"rapl_dram_0_watt": 2,"rapl_package_0_watt": 13,"rapl_total_watt": 15,"sensors_acpi_watt": 26.00},"energizta_version": "0.1a"}

{"interval_us": 5200567,"duration_us": 41604549,"nb_states": 8,"cpu_iowait_pct": 0,"cpu_sys_pct": 9,"cpu_usr_pct": 23,"load1": 2.30,"mem_free_MB": 20896,"mem_total_MB": 32065,"mem_used_MB": 5316,"sda_pct_busy": 1,"sda_read_bps": 2,"sda_write_bps": 19,"sdb_pct_busy": 1,"sdb_read_bps": 2,"sdb_write_bps": 20,"sdc_pct_busy": 1,"sdc_read_bps": 1,"sdc_write_bps": 18,"sensors_coretemp": 52.0,"powers": {"dcmi_cur_watt": 48,"rapl_dram_0_watt": 2,"rapl_package_0_watt": 25,"rapl_total_watt": 27,"sensors_acpi_watt": 39.00},"energizta_version": "0.1a"}

{"interval_us": 5356037,"duration_us": 42848298,"nb_states": 8,"cpu_iowait_pct": 0,"cpu_sys_pct": 9,"cpu_usr_pct": 60,"load1": 6.20,"mem_free_MB": 20837,"mem_total_MB": 32065,"mem_used_MB": 5373,"sda_pct_busy": 2,"sda_read_bps": 2,"sda_write_bps": 17,"sdb_pct_busy": 2,"sdb_read_bps": 3,"sdb_write_bps": 17,"sdc_pct_busy": 2,"sdc_read_bps": 1,"sdc_write_bps": 19,"sensors_coretemp": 63.0,"powers": {"dcmi_cur_watt": 54,"rapl_dram_0_watt": 2,"rapl_package_0_watt": 42,"rapl_total_watt": 44,"sensors_acpi_watt": 55.00},"energizta_version": "0.1a"}

{"interval_us": 6842589,"duration_us": 54740718,"nb_states": 8,"cpu_iowait_pct": 0,"cpu_sys_pct": 9,"cpu_usr_pct": 90,"load1": 10.25,"mem_free_MB": 20781,"mem_total_MB": 32065,"mem_used_MB": 5429,"sda_pct_busy": 2,"sda_read_bps": 0,"sda_write_bps": 18,"sdb_pct_busy": 2,"sdb_read_bps": 1,"sdb_write_bps": 16,"sdc_pct_busy": 2,"sdc_read_bps": 2,"sdc_write_bps": 17,"sensors_coretemp": 65.0,"powers": {"dcmi_cur_watt": 48,"rapl_dram_0_watt": 2,"rapl_package_0_watt": 48,"rapl_total_watt": 50,"sensors_acpi_watt": 65.00},"energizta_version": "0.1a"}