Logging & Runtime

This page documents “general” configuration keys that are not specific to flows/topology.

Implementation notes:

• CsvLogger reads logging keys from src/utils/logger.rs.
• Runtime/UI/tracing keys are read from small config structs in src/topos/topo.rs and src/utils/.

seed (required)

Random seed used for reproducibility (flow-set endpoint selection, random distributions, RED randomness).

seed = 1000

duration (optional)

Total simulated time in seconds.

• Default (topology): 1500.0

duration = 20.0

Note: duration is parsed by multiple subsystems (topology, UI, tracing). For consistency, prefer setting it explicitly.

threading (optional)

Threading model for the simulation runtime.

threading = "single"

• Allowed values: single, multiple
• single uses a single-threaded runtime.
• multiple uses one worker per CPU core by default.
• If omitted, Days uses the nexosim default runtime configuration.

num_threads (optional)

Overrides the number of runtime worker threads when threading = "multiple".

threading = "multiple"
num_threads = 4

hot_workers (optional)

Number of hot standby worker threads. Hot workers remain active and ready to pick up tasks, reducing latency for bursty workloads.

hot_workers = 2

concurrency_level (optional)

Concurrency acceleration mode for the simulation runtime.

• Allowed values: default, accelerated

concurrency_level = "accelerated"

time_quantum_ns (optional)

Simulation time quantum in nanoseconds. Controls the granularity of time-based scheduling in the simulation engine.

time_quantum_ns = 1000

mailbox_capacity (optional)

Mailbox capacity for most models (bounded message queues).

• Default: 16

mailbox_capacity = 32

CSV reporting

log_path (optional)

Output directory for CSV files.

• Default: ./output

log_path = "./logs/run1"

Days will create the directory if needed and will write:

• sources.csv, sinks.csv, switches.csv
• plus optional pfc.csv depending on features.
• plus optional per-transition protocol traces for LeanGuard (*_events.csv) depending on features:
  • aqm_events.csv (--features lean)
  • dcqcn_events.csv (--features dcqcn,lean)
  • pfc_events.csv (--features l2_pfc,lean)
  • cubic_events.csv (--features lean)
  • drr_events.csv (--features lean)
  • wfq_events.csv (--features lean)
• traces.json: a manifest listing which *_events.csv traces are non-empty (written at the end of a run).

report_interval (optional)

Interval in seconds between periodic report rows.

• Default: f64::MAX (effectively “no periodic reports”; data is flushed at the end)

report_interval = 1.0

Progress UI

ui_interval (optional)

UI refresh interval in seconds.

• Default: duration / 100

ui_interval = 1.0

Concurrency tracing (optional)

Enable lightweight concurrency sampling:

tracing_active = true
tracing_interval = 1.0

• tracing_active (default: false)
• tracing_interval (default: duration / 100)

When enabled, Days installs a tracing subscriber layer and computes a wall-clock average concurrency during Simulation::step_until (and logs the peak observed by the tracing layer).

Application-level sources for TCP collectives (optional)

For certain TCP collectives (broadcast, ring allreduce), Days can back TCP flows with a shared, actor-managed byte buffer (src/flows/app_source.rs).

Configure the actor under [app_source]:

[app_source]
req_channel_capacity = 256
chunk_size = 512
initial_delay = 1
run_interval = 50

Fields and defaults:

If [app_source] is omitted entirely, Days uses AppBufferConfig::default() with the defaults above.