docs: optimize architecture diagram layout with vertical flow

- Reorganize diagram to show clear top-to-bottom flow:
  ① vLLM Semantic Router Layer (top)
  ② NVIDIA Dynamo Layer (middle)
  ③ Execution Layer - Worker Pools (bottom)
- Place Storage Layer on the side for better space utilization
- Add numbered labels (①②③) to emphasize layer hierarchy
- Wrap main processing layers in a 'Main Processing Flow' subgraph
- Improve visual clarity with consistent direction indicators
- Update step numbers in flow annotations (1-7)

This layout better reflects the actual request flow from client
through semantic intelligence, infrastructure routing, to execution.

Signed-off-by: bitliu <[email protected]>

Author: Xunzhuo

website/docs/proposals/nvidia-dynamo-integration.md

@@ -519,57 +519,61 @@ prompt_guard:
 graph TB
     Client[LLM Application<br/>OpenAI SDK]
 
-    subgraph SIL["Semantic Intelligence Layer"]
+    subgraph Main["Main Processing Flow"]
         direction TB
-        Gateway[Envoy Gateway :8080]
-        ExtProc[Semantic Router ExtProc :50051]
-
-        subgraph SC["Semantic Components"]
-            direction LR
-            Classifier[BERT Classifier]
-            PIIDetector[PII Detector]
-            JailbreakGuard[Jailbreak Guard]
-        end
 
-        SemanticCache[Semantic Cache]
-        ToolSelector[Tool Selector]
-    end
+        subgraph SIL["① vLLM Semantic Router Layer"]
+            direction TB
+            Gateway[Envoy Gateway :8080]
+            ExtProc[Semantic Router ExtProc :50051]
 
-    subgraph DL["NVIDIA Dynamo Layer"]
-        direction TB
-        DynamoFrontend[Dynamo Frontend :8000]
+            subgraph SC["Semantic Components"]
+                direction LR
+                Classifier[BERT Classifier]
+                PIIDetector[PII Detector]
+                JailbreakGuard[Jailbreak Guard]
+            end
 
-        subgraph DR["Routing & Management"]
-            direction LR
-            DynamoRouter[KV Router]
-            KVBM[KV Block Manager]
+            SemanticCache[Semantic Cache]
+            ToolSelector[Tool Selector]
         end
 
-        Planner[Planner - Dynamic Scaling]
-    end
+        subgraph DL["② NVIDIA Dynamo Layer"]
+            direction TB
+            DynamoFrontend[Dynamo Frontend :8000]
 
-    subgraph EL["Execution Layer - Worker Pools"]
-        direction TB
+            subgraph DR["Routing & Management"]
+                direction LR
+                DynamoRouter[KV Router]
+                KVBM[KV Block Manager]
+            end
 
-        subgraph MP1["deepseek-v31"]
-            direction LR
-            W1[Prefill Worker]
-            W2[Decode Worker]
+            Planner[Planner - Dynamic Scaling]
         end
 
-        subgraph MP2["phi4"]
-            direction LR
-            W3[Prefill Worker]
-            W4[Decode Worker]
-        end
+        subgraph EL["③ Execution Layer - Worker Pools"]
+            direction TB
+
+            subgraph MP1["Model Pool: deepseek-v31"]
+                direction LR
+                W1[Prefill Worker]
+                W2[Decode Worker]
+            end
 
-        subgraph MP3["qwen3"]
-            W5[Worker - SGLang]
+            subgraph MP2["Model Pool: phi4"]
+                direction LR
+                W3[Prefill Worker]
+                W4[Decode Worker]
+            end
+
+            subgraph MP3["Model Pool: qwen3"]
+                W5[Worker - SGLang]
+            end
         end
     end
 
     subgraph SL["Storage Layer"]
-        direction LR
+        direction TB
         Milvus[(Milvus<br/>Semantic Cache)]
         SystemMem[(System Memory<br/>KV Offload)]
         NVMe[(NVMe<br/>Cold Cache)]
@@ -587,30 +591,31 @@ graph TB
     DynamoFrontend --> DynamoRouter
     DynamoRouter <--> KVBM
 
-    DynamoRouter --> W1
-    DynamoRouter --> W2
+    DynamoRouter -->|4. Worker Selection| W1
+    DynamoRouter -->|4. Worker Selection| W2
     DynamoRouter -.-> W3
     DynamoRouter -.-> W4
     DynamoRouter -.-> W5
 
-    Planner -.-> W1
-    Planner -.-> W2
-    Planner -.-> W3
-    Planner -.-> W4
-    Planner -.-> W5
+    Planner -.->|Scaling| W1
+    Planner -.->|Scaling| W2
+    Planner -.->|Scaling| W3
+    Planner -.->|Scaling| W4
+    Planner -.->|Scaling| W5
 
     SemanticCache <--> Milvus
     KVBM <--> SystemMem
     KVBM <--> NVMe
 
-    W1 -->|4. Response| DynamoFrontend
-    DynamoFrontend --> Gateway
-    Gateway --> Client
+    W1 -->|5. Response| DynamoFrontend
+    DynamoFrontend -->|6. Response| Gateway
+    Gateway -->|7. Response| Client
 
     style ExtProc fill:#e1f5ff
     style DynamoRouter fill:#c8e6c9
     style SemanticCache fill:#fff9c4
     style KVBM fill:#fff9c4
+    style SL fill:#f5f5f5
 ```
 
 **Architecture Layers:**