host.parquet

host.parquet has one row per host per sample. It describes the state, utilization and power draw of each host.

Column	Type	Unit	Description
timestamp	int64	ms	The time of the sample.
host_id	int32		The id of the host. To get the name, see meta/host.parquet.
host_state	string		UP, DOWN or ERROR.
tasks_running	int32	count	The number of tasks running on the host.
tasks_error	int32	count	The number of tasks on the host that failed.
tasks_invalid	int32	count	The number of tasks on the host in an unexpected state.
cpu_usage	float	MHz	The CPU capacity used by the tasks.
cpu_demand	float	MHz	The CPU capacity the tasks ask for.
cpu_utilization	float	ratio	cpu_usage divided by the CPU capacity of the host.
cpu_time_active	int64	ms	The total time the CPU was active.
cpu_time_idle	int64	ms	The total time the CPU was idle.
cpu_time_steal	int64	ms	The total time tasks wanted to run, but no CPU capacity was available.
cpu_time_lost	int64	ms	The total CPU time lost to interference.
power_draw	float	W	The power draw of the host.
energy_usage	float	J	The energy the host used since the start of the simulation.
carbon_intensity	float	gCO2/kWh	The carbon intensity of the energy.
carbon_emission	float	g	The carbon emitted since the start of the simulation.
embodied_carbon	float	g	The embodied carbon of the host, for the time since the previous sample.
uptime	int64	ms	The total time the host was up.
downtime	int64	ms	The total time the host was down because of a failure.
boot_time	int64	ms	When the host booted, or null if it has not booted.

GPU columns

When the topology has GPUs, the file has a set of columns for each GPU of a host. The columns of the first GPU end in _0, those of the second in _1, and so on. A host with fewer GPUs than the host with the most GPUs has null values for the missing ones.

Column	Type	Unit	Description
gpu_usage_i	float	MHz	The GPU capacity used by the tasks.
gpu_demand_i	float	MHz	The GPU capacity the tasks ask for.
gpu_utilization_i	float	ratio	gpu_usage_i divided by the GPU capacity.
gpu_time_active_i	int64	ms	The total time the GPU was active.
gpu_time_idle_i	int64	ms	The total time the GPU was idle.
gpu_time_steal_i	int64	ms	The total time tasks wanted to run, but no GPU capacity was available.
gpu_time_lost_i	int64	ms	The total GPU time lost to interference.
gpu_power_draw_i	float	W	The power draw of the GPU.