Concat after chunks moving to ensure usability

Author: linamy85

Ironwood/configs/host_device/host_device_single_chip.yaml

@@ -3,7 +3,11 @@ benchmarks:
   num_runs: 20
   benchmark_sweep_params:
   # Single Chip (1 Chip, 2 Devices)
-  - {mesh_shape: "1x2", data_size_mb_list: [1, 16, 128, 256, 512, 1024, 2048, 4096, 8192, 16384, 32768]}
+  - {
+      num_devices: 2,
+      smart_chunking: False,
+      data_size_mb_list: [1, 16, 128, 256, 512, 1024, 2048, 4096, 8192, 16384, 32768]
+    }
 
   csv_path: "../microbenchmarks/host_device/single_chip"
   trace_dir: "../microbenchmarks/host_device/single_chip/trace"

Ironwood/configs/host_device/host_device_single_chip_smart_chunking.yaml

@@ -0,0 +1,14 @@
+benchmarks:
+- benchmark_name: host_device
+  num_runs: 20
+  benchmark_sweep_params:
+  # Single Chip (1 Chip, 2 Devices)
+  # Data sizes: 1MB to 32GB (powers of 2)
+  - {
+      num_devices: 2, 
+      smart_chunking: True,
+      data_size_mb_list: [1, 16, 128, 256, 512, 1024, 2048, 4096, 8192, 16384, 32768]
+    }
+
+  csv_path: "../microbenchmarks/host_device/single_chip_smart_chunking"
+  trace_dir: "../microbenchmarks/host_device/single_chip_smart_chunking/trace"

Ironwood/guides/host_device/host_device.md

@@ -28,3 +28,16 @@ To retrieve the complete results, including the trace and CSV output files, you
 ```bash
 kubectl cp tpu7x-host-device-benchmark:/microbenchmarks/host_device host_device
 ```
+
+## Run Smart Chunking Mode (Optional)
+
+To run the benchmark with "Smart Chunking" enabled (searching for optimal transfer chunk size):
+
+```bash
+kubectl apply -f tpu7x-host-device-benchmark-smart-chunking.yaml
+```
+
+View logs:
+```bash
+kubectl logs tpu7x-host-device-benchmark-smart-chunking
+```

Ironwood/guides/host_device/tpu7x-host-device-benchmark-smart-chunking.yaml

@@ -0,0 +1,34 @@
+apiVersion: v1
+kind: Pod
+metadata:
+  name: tpu7x-host-device-benchmark-smart-chunking
+spec:
+  restartPolicy: Never
+  nodeSelector:
+    cloud.google.com/gke-tpu-accelerator: tpu7x
+    cloud.google.com/gke-tpu-topology: 2x2x1
+  containers:
+  - name: tpu-job
+    image: python:3.12
+    ports:
+    - containerPort: 8431
+    securityContext:
+      privileged: false
+    command:
+    - bash
+    - -c
+    - |
+      set -ex
+
+      git clone https://github.com/AI-Hypercomputer/accelerator-microbenchmarks.git
+      cd accelerator-microbenchmarks
+      pip install -r requirements.txt
+
+      # Run ONLY the smart chunking config
+      bash ./Ironwood/scripts/run_host_device_benchmark.sh --config Ironwood/configs/host_device/host_device_single_chip_smart_chunking.yaml
+
+    resources:
+      requests:
+        google.com/tpu: 4
+      limits:
+        google.com/tpu: 4

Ironwood/guides/host_device/tpu7x-host-device-benchmark.yaml

@@ -24,7 +24,7 @@ spec:
       cd accelerator-microbenchmarks
       pip install -r requirements.txt
 
-      bash ./Ironwood/scripts/run_host_device_benchmark.sh
+      bash ./Ironwood/scripts/run_host_device_benchmark.sh --config Ironwood/configs/host_device/host_device_single_chip.yaml
 
     resources:
       requests:

Ironwood/scripts/run_host_device_benchmark.sh

@@ -27,7 +27,10 @@ while [[ "$#" -gt 0 ]]; do
 done
 
 echo "--- Starting Host-Device Transfer Benchmark (H2D/D2H) ---"
-echo "Note: This benchmark is work in progress"
+echo "*****************************************************"
+echo "*  WARNING: THIS BENCHMARK IS A WORK IN PROGRESS    *"
+echo "*  Results may be unstable or subject to change.    *"
+echo "*****************************************************"
 echo "Interleaved: $INTERLEAVED"
 
 if [ -n "$SPECIFIC_CONFIG" ]; then