@@ -53,6 +53,10 @@ Once the job starts running, it will hold waiting for the data stream. To start
python sender.py
```
### Changing the Scale of the Benchmark
The benchmark’s scale can be adjusted by altering the number of replicas in the workflow execution. At its smallest scale, the benchmark utilizes 9 nodes (a single replica). To increase the number of replicas, modify the `job.sb` submission script by updating the node count and the number of replicas. For example, to use 900 nodes and 100 replicas, adjust the script to `#SBATCH -N 900` and set `REPLICAS=100` accordingly.
## Run Rules
- _Weak Scaling Experiments:_ Each rank at level 1 (refer to Figure 2) trains a TFT model on 64 ([4, 4, 4]) voxels, and the level 2 rank operates on [2, 2, 2] mean voxels of level 1. Consequently, for an input with dimensions 8 × 8 × 8, a total of 9 ranks (eight in level 1 and one in level 2) are required. For a larger input of 64 × 64 × 64, a total of 4608 ranks are needed, divided into 4096 in level 1 and 512 in level 2.
@@ -70,10 +74,6 @@ The primary figure of merit for the ML4NSE (defined in detail in https://doi.org
An additional potential figure of merit that could demonstrate the robustness of the system would include any active guidance between the Compute and Services Clusters; the latency involved in control operations becomes crucial. Specifically, it's essential to assess the duration a compute job is held while disseminating new control information. Also, as additional input streams and output consumers are added, the effect on end-to-end time-to-solution could be affected.
