Ansys Fluent Parallel Solver: Re-Partitioning Triggers When Dynamic Mesh Updates Shift Load Balance

Answer

The automatic load‑balancing in Ansys Fluent (2023 R1 and later) is triggered when the cell‑count imbalance between the most‑loaded and least‑loaded partition exceeds a configurable threshold and a minimum time‑step interval has elapsed.

Confirmed facts

• Default imbalance threshold: 0.10 (10 % difference).
• Default minimum interval: 5 time steps or 0.5 s of simulated time, whichever is larger.
• These values are exposed via the TUI (/solve/parallel/load-balance/set-interval and /solve/parallel/load-balance/set-imbalance-threshold) and the GUI under Parallel → Partition → Load Balance Settings.

Likely explanation of the trigger

After each dynamic‑mesh update Fluent evaluates the current partition imbalance. If the imbalance > threshold **and** the minimum interval since the last repartition has passed, Fluent launches a repartitioning step that redistributes cells, updates ghost layers and resets the interval timer.

CPU‑only vs GPU‑accelerated runs

The decision logic uses only partition cell counts, so it is identical for CPU‑only and GPU‑accelerated solver configurations. In practice GPU runs often show less sensitivity to imbalance because the compute kernel hides communication latency; many users therefore raise the threshold (e.g., to 15‑20 %) to avoid unnecessary repartitioning overhead.

Steps to verify or adjust the settings

1. Open the TUI and query current values:

   (solve/parallel/load-balance/get-imbalance-threshold)

   (solve/parallel/load-balance/get-interval)

2. If you wish to change them, use:

   /solve/parallel/load-balance/set-imbalance-threshold 0.15

   /solve/parallel/load-balance/set-interval 10

3. Enable tracing to see when repartitioning occurs:

   /solve/parallel/load-balance/trace on

   then run a transient dynamic‑mesh case and inspect the console for messages containing “load balance” and the reported imbalance percentage.
4. Compare the reported imbalance with the threshold; if repartitioning happens far more often than expected, consider raising the threshold or increasing the interval.

Missing diagnostic detail

Please share the observed imbalance percentage from the load‑balance trace (or the console message) so we can confirm whether the current threshold is being exceeded and whether adjusting it is advisable.