Agentic workload PD separation analysis with trace-driven benchmarks

Systematic study of prefill-decode disaggregation for agentic LLM workloads
using production GLM-5.1 coder trace (2.1M requests, 71B input tokens).

Key findings:
- Cache-aware routing improves TPOT p90 by 15% and APC from 20.8% to 44.7%
  without PD separation, matching PD-Sep's decode isolation benefit
- PD separation adds +72% TTFT overhead (KV transfer) with no TPOT gain
  when using the same cache-aware scheduler
- Prefill remains compute-bound even at 95% KV cache reuse (AI >1000x
  vs decode AI <2), but absolute FLOPs drop 71% from cache hits
- For agentic MoE workloads, cache-aware routing > PD separation

Infrastructure:
- Trace sampler preserving session structure + hash_ids for prefix sharing
- Async trace replayer with streaming TTFT/TPOT/E2E measurement
- Unified cache-aware + token-level load-balanced global scheduler proxy
  supporting both PD-colocated and PD-disaggregated (Mooncake/RDMA) modes
- vLLM 0.18.1 scheduler patch for KV transfer abort race condition
- Roofline analysis tool for prefill/decode compute characterization

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

scripts/launch_vllm.sh

@@ -0,0 +1,23 @@
+#!/bin/bash
+# Launch vLLM 0.18.1 in PD-combined mode (TP=8, all GPUs).
+#
+# Usage: bash scripts/launch_vllm.sh
+
+set -euo pipefail
+
+SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+PROJECT_DIR="$(dirname "$SCRIPT_DIR")"
+VLLM="$PROJECT_DIR/.venv/bin/vllm"
+
+MODEL_PATH="${MODEL_PATH:-$HOME/models/Qwen/Qwen3-Coder-30B-A3B-Instruct}"
+HOST="${HOST:-0.0.0.0}"
+PORT="${PORT:-8000}"
+
+echo "Starting vLLM 0.18.1 in PD-combined mode (TP=8) on port $PORT ..."
+$VLLM serve "$MODEL_PATH" \
+    --trust-remote-code \
+    --enable-prefix-caching \
+    --dtype auto \
+    --tensor-parallel-size 8 \
+    --host "$HOST" \
+    --port "$PORT"