Add session affinity perf configurations and summary

test/perf/config/router-configs/load-aware-session-affinity-2rep.yaml

@@ -0,0 +1,78 @@
+# Router configuration that combines load-aware-scorer with session-affinity-filter and passthrough-parser.
+# Configured with 2 active-active EPP replicas for horizontal scaling benchmark.
+router:
+  extraServicePorts:
+    - name: http
+      port: 80
+      protocol: TCP
+      targetPort: 8081
+
+  epp:
+    replicas: 2
+    image:
+      registry: ghcr.io/llm-d
+      repository: llm-d-router-endpoint-picker
+      tag: main
+      pullPolicy: Always
+    flags:
+      v: 4
+      enable-pprof: "true"
+      tracing: "false"
+      ha-enable-leader-election: "false"
+    pluginsConfigFile: "optimized-baseline-plugins.yaml"
+    pluginsCustomConfig:
+      optimized-baseline-plugins.yaml: |
+        apiVersion: llm-d.ai/v1alpha1
+        kind: EndpointPickerConfig
+        plugins:
+        - type: passthrough-parser
+        - type: session-affinity-filter
+          parameters:
+            headerName: x-session-token
+        - type: load-aware-scorer
+          parameters:
+            threshold: 128
+        - type: weighted-random-picker
+        requestHandler:
+          parsers:
+          - pluginRef: passthrough-parser
+        schedulingProfiles:
+        - name: default
+          plugins:
+          - pluginRef: session-affinity-filter
+          - pluginRef: load-aware-scorer
+            weight: 5
+          - pluginRef: weighted-random-picker
+
+  inferencePool:
+    failureMode: "FailOpen"
+
+  modelServers:
+    protocol: http
+
+  proxy:
+    args:
+      - "--service-node"
+      - "envoy-sidecar"
+      - "--log-level"
+      - "warn"
+      - "--concurrency"
+      - "32"
+      - "--drain-strategy"
+      - "immediate"
+      - "--drain-time-s"
+      - "60"
+      - "-c"
+      - "/etc/envoy/envoy.yaml"
+    resources:
+      requests:
+        cpu: "4"
+        memory: 8Gi
+      limits:
+        memory: 16Gi
+
+  tracing:
+    otelExporterEndpoint: "http://localhost:4317"
+    sampling:
+      sampler: "parentbased_traceidratio"
+      samplerArg: "0.1"

test/perf/config/router-configs/load-aware-session-affinity-3rep.yaml

@@ -0,0 +1,78 @@
+# Router configuration that combines load-aware-scorer with session-affinity-filter and passthrough-parser.
+# Configured with 3 active-active EPP replicas for horizontal scaling benchmark.
+router:
+  extraServicePorts:
+    - name: http
+      port: 80
+      protocol: TCP
+      targetPort: 8081
+
+  epp:
+    replicas: 3
+    image:
+      registry: ghcr.io/llm-d
+      repository: llm-d-router-endpoint-picker
+      tag: main
+      pullPolicy: Always
+    flags:
+      v: 4
+      enable-pprof: "true"
+      tracing: "false"
+      ha-enable-leader-election: "false"
+    pluginsConfigFile: "optimized-baseline-plugins.yaml"
+    pluginsCustomConfig:
+      optimized-baseline-plugins.yaml: |
+        apiVersion: llm-d.ai/v1alpha1
+        kind: EndpointPickerConfig
+        plugins:
+        - type: passthrough-parser
+        - type: session-affinity-filter
+          parameters:
+            headerName: x-session-token
+        - type: load-aware-scorer
+          parameters:
+            threshold: 128
+        - type: weighted-random-picker
+        requestHandler:
+          parsers:
+          - pluginRef: passthrough-parser
+        schedulingProfiles:
+        - name: default
+          plugins:
+          - pluginRef: session-affinity-filter
+          - pluginRef: load-aware-scorer
+            weight: 5
+          - pluginRef: weighted-random-picker
+
+  inferencePool:
+    failureMode: "FailOpen"
+
+  modelServers:
+    protocol: http
+
+  proxy:
+    args:
+      - "--service-node"
+      - "envoy-sidecar"
+      - "--log-level"
+      - "warn"
+      - "--concurrency"
+      - "32"
+      - "--drain-strategy"
+      - "immediate"
+      - "--drain-time-s"
+      - "60"
+      - "-c"
+      - "/etc/envoy/envoy.yaml"
+    resources:
+      requests:
+        cpu: "4"
+        memory: 8Gi
+      limits:
+        memory: 16Gi
+
+  tracing:
+    otelExporterEndpoint: "http://localhost:4317"
+    sampling:
+      sampler: "parentbased_traceidratio"
+      samplerArg: "0.1"

test/perf/config/router-configs/load-aware-session-affinity.yaml

@@ -0,0 +1,76 @@
+# Router configuration that combines load-aware-scorer with session-affinity-filter and passthrough-parser.
+router:
+  extraServicePorts:
+    - name: http
+      port: 80
+      protocol: TCP
+      targetPort: 8081
+
+  epp:
+    replicas: 1
+    image:
+      registry: ghcr.io/llm-d
+      repository: llm-d-router-endpoint-picker
+      tag: main
+      pullPolicy: Always
+    flags:
+      v: 4
+      enable-pprof: "true"
+      tracing: "false"
+    pluginsConfigFile: "optimized-baseline-plugins.yaml"
+    pluginsCustomConfig:
+      optimized-baseline-plugins.yaml: |
+        apiVersion: llm-d.ai/v1alpha1
+        kind: EndpointPickerConfig
+        plugins:
+        - type: passthrough-parser
+        - type: session-affinity-filter
+          parameters:
+            headerName: x-session-token
+        - type: load-aware-scorer
+          parameters:
+            threshold: 128
+        - type: weighted-random-picker
+        requestHandler:
+          parsers:
+          - pluginRef: passthrough-parser
+        schedulingProfiles:
+        - name: default
+          plugins:
+          - pluginRef: session-affinity-filter
+          - pluginRef: load-aware-scorer
+            weight: 5
+          - pluginRef: weighted-random-picker
+
+  inferencePool:
+    failureMode: "FailOpen"
+
+  modelServers:
+    protocol: http
+
+  proxy:
+    args:
+      - "--service-node"
+      - "envoy-sidecar"
+      - "--log-level"
+      - "warn"
+      - "--concurrency"
+      - "32"
+      - "--drain-strategy"
+      - "immediate"
+      - "--drain-time-s"
+      - "60"
+      - "-c"
+      - "/etc/envoy/envoy.yaml"
+    resources:
+      requests:
+        cpu: "4"
+        memory: 8Gi
+      limits:
+        memory: 16Gi
+
+  tracing:
+    otelExporterEndpoint: "http://localhost:4317"
+    sampling:
+      sampler: "parentbased_traceidratio"
+      samplerArg: "0.1"

test/perf/config/router-configs/optimized-baseline.yaml

@@ -56,7 +56,7 @@ router:
       - "--log-level"
       - "warn"
       - "--concurrency"
-      - "8"
+      - "32"
       - "--drain-strategy"
       - "immediate"
       - "--drain-time-s"

test/perf/config/router-configs/pd-session-affinity.yaml

@@ -0,0 +1,87 @@
+# Router configuration for Prefill/Decode (P/D) disaggregation with session affinity using llm-d-sim.
+router:
+  extraServicePorts:
+    - name: http
+      port: 80
+      protocol: TCP
+      targetPort: 8081
+
+  epp:
+    replicas: 1
+    image:
+      registry: ghcr.io/llm-d
+      repository: llm-d-router-endpoint-picker
+      tag: main
+      pullPolicy: Always
+    flags:
+      v: 4
+      enable-pprof: "true"
+      tracing: "false"
+      ha-enable-leader-election: "false"
+    pluginsConfigFile: "optimized-baseline-plugins.yaml"
+    pluginsCustomConfig:
+      optimized-baseline-plugins.yaml: |
+        apiVersion: llm-d.ai/v1alpha1
+        kind: EndpointPickerConfig
+        plugins:
+        - type: passthrough-parser
+        - type: always-disagg-pd-decider
+        - type: disagg-profile-handler
+          parameters:
+            deciders:
+              prefill: always-disagg-pd-decider
+        - type: session-affinity-filter
+          name: session-affinity-decode
+          parameters:
+            headerName: x-session-token
+        - type: session-affinity-filter
+          name: session-affinity-prefill
+          parameters:
+            headerName: x-session-token-prefill
+            profileName: prefill
+        - type: random-picker
+        requestHandler:
+          parsers:
+          - pluginRef: passthrough-parser
+        schedulingProfiles:
+        - name: prefill
+          plugins:
+          - pluginRef: session-affinity-prefill
+          - pluginRef: random-picker
+        - name: decode
+          plugins:
+          - pluginRef: session-affinity-decode
+          - pluginRef: random-picker
+
+  inferencePool:
+    failureMode: "FailOpen"
+
+  modelServers:
+    protocol: http
+
+  proxy:
+    args:
+      - "--service-node"
+      - "envoy-sidecar"
+      - "--log-level"
+      - "warn"
+      - "--concurrency"
+      - "32"
+      - "--drain-strategy"
+      - "immediate"
+      - "--drain-time-s"
+      - "60"
+      - "-c"
+      - "/etc/envoy/envoy.yaml"
+    resources:
+      requests:
+        cpu: "4"
+        memory: 8Gi
+      limits:
+        memory: 16Gi
+
+  tracing:
+    otelExporterEndpoint: "http://localhost:4317"
+    sampling:
+      sampler: "parentbased_traceidratio"
+      samplerArg: "0.1"