Struct TraceThroughputStats 

pub struct TraceThroughputStats {
    pub duration_ms: f64,
    pub wall_time_ms: f64,
    pub request_throughput_rps: f64,
    pub input_throughput_tok_s: f64,
    pub output_throughput_tok_s: f64,
    pub total_throughput_tok_s: f64,
    pub prefill_worker_seconds: f64,
    pub decode_worker_seconds: f64,
    pub prefill_gpus_per_worker: usize,
    pub decode_gpus_per_worker: usize,
    pub gpu_hours: f64,
}

Fields

duration_ms: f64

wall_time_ms: f64

request_throughput_rps: f64

input_throughput_tok_s: f64

output_throughput_tok_s: f64

total_throughput_tok_s: f64

prefill_worker_seconds: f64

Provisioned worker-time per role, in worker-seconds: the time-integral of the provisioned worker count over the whole simulated run. The provisioned count is every worker physically holding a GPU (active + starting-up + draining), so this captures the startup ramp and the scale-down drain tail, unlike a snapshot of the active/serving count. Populated on the collector by the runtime: add_worker_seconds accrues the integral each clock advance (agg / disagg), and set_static_worker_count covers the single-worker path; 0.0 otherwise. Multiply by GPUs-per-worker for GPU-seconds (/3600 for GPU-hours). Aggregated replay reports through decode_worker_seconds, leaving prefill_worker_seconds at 0.0.

decode_worker_seconds: f64

prefill_gpus_per_worker: usize

GPUs per worker per role, derived from the mocker engine parallelism (MockEngineArgs::aic_gpus_per_worker = tensor parallelism × materialized DP topology); the runtime sets it on the collector. 0 when not set.

decode_gpus_per_worker: usize

gpu_hours: f64

GPU-hours = Σ_role worker_seconds × gpus_per_worker / 3600 — the deployment’s provisioned GPU-time (already including the startup ramp and drain tail, since *_worker_seconds do). Computed in finish() straight from the mocker’s own worker parallelism, so it needs no external config.

Trait Implementations

impl Clone for TraceThroughputStats

fn clone(&self) -> TraceThroughputStats

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for TraceThroughputStats

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.