Gahow Wang

gahow pushed to main at gahow/agentic-kvc

2026-05-27 14:25:00 +00:00

4a93096c1e Add PD_DISAGG_INVESTIGATION.md — living TODO for proving H1–H4

gahow pushed to main at gahow/agentic-kvc

2026-05-27 14:05:29 +00:00

f739f7d461 Proxy/runner support for Nixl connector + unified_v3 (offload-decode) policy

da39ab6804 Correct PD-disagg cost/benefit framing across repo

Compare 2 commits »

gahow pushed to main at gahow/agentic-kvc

2026-05-27 14:02:57 +00:00

4936c6b861 Correct PD-disagg cost/benefit framing across repo

gahow pushed to main at gahow/agentic-kvc

2026-05-27 13:38:42 +00:00

abde010b64 Add RESULTS_SUMMARY.md — concise Chinese summary of current findings

gahow pushed to main at gahow/agentic-kvc

2026-05-27 13:25:12 +00:00

029821c1b6 MB1: prefill-decode interference under chunked-prefill default; §3.2 headline

gahow pushed to main at gahow/agentic-kvc

2026-05-27 12:56:13 +00:00

90127c3389 MB2 inter-node: dash1↔dash2 transfer cost is identical to intra-node

gahow pushed to main at gahow/agentic-kvc

2026-05-27 12:27:03 +00:00

50f72d8875 MB2 inter-node scaffolding: per-host single-instance launcher + client host args

gahow pushed to main at gahow/agentic-kvc

2026-05-27 12:24:00 +00:00

3f791ee074 MB2 doc: analysis/mb2/README.md as persistent record

gahow pushed to main at gahow/agentic-kvc

2026-05-27 11:04:07 +00:00

de164e5a64 MB2: pure KV-transfer cost on dash1 intra-node — Mooncake ~9.7 GB/s steady

gahow pushed to main at gahow/agentic-kvc

2026-05-27 10:53:29 +00:00

91673f1fb8 MB2: working end-to-end intra-node KV transfer microbench

gahow pushed to main at gahow/agentic-kvc

2026-05-27 10:17:46 +00:00

622e0bc04c MB2: parameterize vLLM roles (kv_producer + kv_consumer default)

gahow pushed to main at gahow/agentic-kvc

2026-05-27 10:12:48 +00:00

efdcf3c555 MB2: per-stage instrumentation patch + launcher integration

gahow pushed to main at gahow/agentic-kvc

2026-05-27 09:47:07 +00:00

7437422618 MB2 scaffolding: launch script for vLLM pair + KV-transfer-time client

gahow pushed to main at gahow/agentic-kvc

2026-05-27 09:42:41 +00:00

0a63de5bcf Phase 0: fresh vllm 0.18.1 + mooncake-transfer-engine on dash1/dash2

gahow pushed to main at gahow/agentic-kvc

2026-05-27 08:51:43 +00:00

b11dc30945 §2.3 reframe: dispatch coupling is regime-dependent, not binary chatbot/agentic

gahow pushed to main at gahow/agentic-kvc

2026-05-27 06:49:50 +00:00

876d09db83 Add chatbot T_external CDF; overlay on f3a vs agentic

gahow pushed to main at gahow/agentic-kvc

2026-05-27 06:04:18 +00:00

cef914ecd4 §3.1: add LMetric vs load_only design analysis (cache signal diluted by ×score)

gahow pushed to main at gahow/agentic-kvc

2026-05-27 05:55:12 +00:00

c33c825256 figs/v2: drop unified_v2 (buggy variant); re-render 4-policy panels

gahow pushed to main at gahow/agentic-kvc

2026-05-27 05:52:21 +00:00

03d8c5d0d1 Render 4 per-policy figures on b3_replay_20260527_0114 into figs/v2/

gahow pushed to main at gahow/agentic-kvc

2026-05-27 04:37:36 +00:00

41232f49d3 Measure inter-turn T_external on the raw production trace; add f3a CDF