KVCache simulator for LLM serving cluster routing research

Discrete-event simulator for evaluating KV cache-aware routing policies
in prefill-disaggregated LLM serving clusters. Models a two-tier KV cache
hierarchy (L0 GPU HBM + L1 CPU DRAM) with RDMA/PCIe link contention,
architecture-derived roofline compute (MoE, MLA, DSA), and a cluster-wide
meta-store for prefix-aware routing decisions.

Includes 11 routing policies (random, round_robin, least_loaded,
least_tokens, ttl_aware, precise, min_pd, cache_load, cache_score,
estimated_ttft, prefix_affinity), HuggingFace config.json auto-parsing,
built-in GPU hardware presets (H100/H800/H20/A100/B200), and ablation
tooling for systematic policy comparison across real Alibaba serving traces.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

src/metrics/per_request.rs

@@ -0,0 +1,42 @@
+use anyhow::Result;
+use serde::Serialize;
+use std::path::Path;
+
+#[derive(Debug, Clone, Serialize)]
+pub struct PerRequestRow {
+    pub req_id: u64,
+    pub arrival: f64,
+    pub ttft: f64,
+    pub e2e: f64,
+    pub instance: u32,
+    pub total_blocks: u32,
+    pub l0_hit_blocks: u32,
+    pub l1_hit_blocks: u32,
+    pub remote_hit_blocks: u32,
+    pub miss_blocks: u32,
+    pub rdma_bytes: u64,
+    pub pcie_bytes: u64,
+    pub probe_overhead_s: f64,
+}
+
+pub struct PerRequestWriter {
+    inner: csv::Writer<std::fs::File>,
+}
+
+impl PerRequestWriter {
+    pub fn create<P: AsRef<Path>>(path: P) -> Result<Self> {
+        let f = std::fs::File::create(path)?;
+        let inner = csv::Writer::from_writer(f);
+        Ok(Self { inner })
+    }
+
+    pub fn write(&mut self, row: &PerRequestRow) -> Result<()> {
+        self.inner.serialize(row)?;
+        Ok(())
+    }
+
+    pub fn finish(mut self) -> Result<()> {
+        self.inner.flush()?;
+        Ok(())
+    }
+}