chore: port basic settings from gb200 disagg

Author: ivanium

benchmarks/benchmark_lib.sh

@@ -1794,7 +1794,11 @@ build_replay_cmd() {
     # X-Correlation-ID is useful tracing metadata but does not establish that
     # binding by itself. AIPerf emits nvext.session_control bind/close actions
     # keyed by the stable conversation correlation ID when this flag is set.
-    if [[ "${FRAMEWORK:-}" == dynamo-* ]]; then
+    # Opt-out: recipes set AIPERF_USE_DYNAMO_CONV_AWARE_ROUTING=0 to skip this.
+    # aiperf's conv-aware routing emits nvext.session_control, a removed POC field
+    # (dynamo #9920 / v1.3.0-dev) that current dynamo builds reject with a 400
+    # (they moved to router/routing_constraints/agent_context). Default stays on.
+    if [[ "${FRAMEWORK:-}" == dynamo-* && "${AIPERF_USE_DYNAMO_CONV_AWARE_ROUTING:-1}" != "0" ]]; then
         REPLAY_CMD+=" --use-dynamo-conv-aware-routing"
         # The upstream 300s affinity TTL is shorter than an overloaded
         # high-concurrency agentic request. Keep bindings alive across long

utils/matrix_logic/generate_sweep_configs.py

@@ -24,7 +24,9 @@
 
 MIN_EVAL_CONC = 16
 # Bound how many multinode agentic conc points share one server allocation.
-MAX_MULTINODE_AGENTIC_CONCURRENCIES_PER_ALLOCATION = 4
+# One task/Slurm allocation per concurrency gives each benchmark point a
+# fresh server deployment, matching single-node agentic sweep isolation.
+MAX_MULTINODE_AGENTIC_CONCURRENCIES_PER_ALLOCATION = 1
 BYTES_PER_MIB = 1024 * 1024
 BYTES_PER_GB = 1_000_000_000
 # 3 TB decimal DRAM cap, expressed in MiB, before utilization scaling.

utils/matrix_logic/test_generate_sweep_configs.py

@@ -2253,8 +2253,8 @@ def test_agentic_node_dram_rejects_tp_above_runner_gpus(self, sample_runner_conf
         with pytest.raises(ValueError, match="exceeds gpus-per-node"):
             generate_test_config_sweep(args, config, runner_config)
 
-    def test_multinode_agentic_groups_concurrencies_per_search_entry(self):
-        """One server allocation should run the selected concurrency batch."""
+    def test_multinode_agentic_uses_one_allocation_per_concurrency(self):
+        """Each concurrency should get its own server allocation."""
         config = {
             "dsv4-agentic-2p1d": {
                 "image": "vllm/vllm-openai:v0.23.0",
@@ -2292,17 +2292,27 @@ def test_multinode_agentic_groups_concurrencies_per_search_entry(self):
 
         result = generate_test_config_sweep(args, config)
 
-        assert len(result) == 2
-        assert result[0]["conc"] == [16, 32, 64, 128]
-        assert result[0]["exp-name"] == "dsv4_p2x4_d1x4_conc16x32x64x128"
+        assert len(result) == 5
+        assert [entry["conc"] for entry in result] == [
+            [16],
+            [32],
+            [64],
+            [128],
+            [256],
+        ]
+        assert [entry["exp-name"] for entry in result] == [
+            "dsv4_p2x4_d1x4_conc16",
+            "dsv4_p2x4_d1x4_conc32",
+            "dsv4_p2x4_d1x4_conc64",
+            "dsv4_p2x4_d1x4_conc128",
+            "dsv4_p2x4_d1x4_conc256",
+        ]
         assert result[0]["prefill"]["pp"] == 2
         assert result[0]["prefill"]["dcp-size"] == 2
         assert result[0]["prefill"]["pcp-size"] == 2
         assert result[0]["decode"]["pp"] == 2
         assert result[0]["decode"]["dcp-size"] == 2
         assert result[0]["decode"]["pcp-size"] == 1
-        assert result[1]["conc"] == [256]
-        assert result[1]["exp-name"] == "dsv4_p2x4_d1x4_conc256"
         assert all(entry["router"] == {"name": "dynamo-router", "version": "1.3.0"} for entry in result)
         assert all(entry["kv-p2p-transfer"] == "nixl" for entry in result)
 
@@ -2493,9 +2503,11 @@ def test_node_type_filters_apply_to_agentic_configs(
         assert "prefill" not in single_result[0]
         assert single_result[0]["runner"] == "cluster:b300-nv"
         assert single_result[0]["pp"] == 2
-        assert len(multi_result) == 1
-        assert "prefill" in multi_result[0]
+        assert len(multi_result) == 2
+        assert [entry["conc"] for entry in multi_result] == [[16], [32]]
+        assert all("prefill" in entry for entry in multi_result)
         assert multi_result[0]["runner"] == "cluster:gb200-nv"
+        assert multi_result[1]["runner"] == multi_result[0]["runner"]
         assert (
             multi_result[0]["prefill"]["pp"],
             multi_result[0]["prefill"]["dcp-size"],