600s-truncated trace + LPWL 5-policy results

traces/w600_r0.0015_st30_first600s.jsonl: first-600s cut of the shipped w600
trace (807 reqs, 274 sessions, all turn-1s + early later-turns; theoretical
APC ceiling ~70% vs 80% full). Faster iteration (~18 min/arm) but a colder,
lower-locality regime; whitelisted alongside the parent anonymized trace.

analysis/lpwl_5policy_600s.md: LPWL vs LMetric/sticky/unified/unified+A+B on
the 600s trace (dash1 8xH20, cold APC, n=1). LPWL is overall best with zero
knobs — TTFT p90 7983ms vs tuned A+B 11562 (-31%), E2E p90 -16%, best request
balance; APC 0.648 (emergent affinity, far above LMetric 0.507); only loss is
E2E p99 from heavy-class decode concentration. Demonstrates anti-overfit: A+B
was tuned on full w600 yet is beaten by the knob-free policy on this regime.
Includes the run_5policy_600s.sh repro driver.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
This commit is contained in:
2026-05-29 16:08:35 +08:00
parent 160c29133d
commit 71b0747b3b
4 changed files with 948 additions and 0 deletions

1
.gitignore vendored
View File

@@ -10,3 +10,4 @@ traces/*
*.log
.claude/
# third_party/vllm tracked in git for patch management
!traces/w600_r0.0015_st30_first600s.jsonl