{ "schema": "anvil-investigation.swift-host-cache-arm/v1", "arm": "control", "server_instance_id": "[operator identity redacted]", "request_accounting": { "preflight_observations": 25, "request_start_count": 43, "request_done_count": 43, "expected_total": 43, "unexpected_or_unmatched": false, "pairing": "IDs1..25 preflight; IDs26..43 exact ordered actual prompt/completion counts match three6-request artifacts; no interpass inference.", "preflight_prompt_completion_multiset_matches": true, "crosspass_prompt_token_sequence_matches": true }, "engine": { "kv_capacity": 282112, "kv_cache": "k8v4", "context_cache": { "device_state_slots": 4, "enabled": true, "host_kv_capacity_bytes": 2147483648, "host_state_slots": 4, "max_long_anchors_per_continuation": 2, "max_private_continuations": 8, "max_shared_prefixes": 4, "total_device_state_slots": 8 }, "max_context": 262144, "max_concurrency": 4, "prefill_chunk": 4096, "speculative_backend": "dflash2", "speculative_draft_window": 7 }, "passes": [ { "pass": "cold", "requests": [ { "request_index": 0, "native_request_id": 26, "prompt_tokens": 62327, "completion_tokens": 302, "prefix_cache_hit_tokens": 0, "computed_prefill_tokens": 62327, "prefix_reuse_path": "root", "visible_ttft_ms": 8976.84140000274, "e2e_ms": 9085.699700000987 }, { "request_index": 1, "native_request_id": 27, "prompt_tokens": 62328, "completion_tokens": 243, "prefix_cache_hit_tokens": 0, "computed_prefill_tokens": 62328, "prefix_reuse_path": "root", "visible_ttft_ms": 8603.36290000123, "e2e_ms": 8711.974600002577 }, { "request_index": 2, "native_request_id": 28, "prompt_tokens": 62328, "completion_tokens": 340, "prefix_cache_hit_tokens": 0, "computed_prefill_tokens": 62328, "prefix_reuse_path": "root", "visible_ttft_ms": 9222.088800001075, "e2e_ms": 9402.902300000278 }, { "request_index": 3, "native_request_id": 29, "prompt_tokens": 62330, "completion_tokens": 194, "prefix_cache_hit_tokens": 0, "computed_prefill_tokens": 62330, "prefix_reuse_path": "root", "visible_ttft_ms": 8442.63940000019, "e2e_ms": 8587.329400001181 }, { "request_index": 4, "native_request_id": 30, "prompt_tokens": 62333, "completion_tokens": 178, "prefix_cache_hit_tokens": 0, "computed_prefill_tokens": 62333, "prefix_reuse_path": "root", "visible_ttft_ms": 8384.973300002457, "e2e_ms": 8475.547300000471 }, { "request_index": 5, "native_request_id": 31, "prompt_tokens": 62335, "completion_tokens": 300, "prefix_cache_hit_tokens": 0, "computed_prefill_tokens": 62335, "prefix_reuse_path": "root", "visible_ttft_ms": 9027.16010000222, "e2e_ms": 9099.811000000045 } ], "median_visible_ttft_ms": 8790.102150001985, "median_e2e_ms": 8898.837150001782, "reused_prompt_tokens": 0, "new_throughput_intervals": 14, "interval_counter_sums": { "actual_transfer_seconds": 0.09356745699999999, "backend_kv_transfers.d2d.bytes": 0, "backend_kv_transfers.d2d.pages": 0, "backend_kv_transfers.d2d.seconds": 0.0, "backend_kv_transfers.d2h.bytes": 0, "backend_kv_transfers.d2h.pages": 0, "backend_kv_transfers.d2h.seconds": 0.0, "backend_kv_transfers.h2d.bytes": 0, "backend_kv_transfers.h2d.pages": 0, "backend_kv_transfers.h2d.seconds": 0.0, "captures.aborted": 0, "captures.completed": 9, "main_kv_transfers.d2d.bytes": 21405696, "main_kv_transfers.d2d.pages": 13, "main_kv_transfers.d2d.seconds": 0.004880736, "main_kv_transfers.d2h.bytes": 1157554176, "main_kv_transfers.d2h.pages": 703, "main_kv_transfers.d2h.seconds": 0.020355713, "main_kv_transfers.h2d.bytes": 0, "main_kv_transfers.h2d.pages": 0, "main_kv_transfers.h2d.seconds": 0.0, "pressure.checkpoints_dropped": 18, "pressure.historical_fork_hits": 9, "pressure.maximal_fallback_selections": 0, "pressure.partial_tail_cow_pages": 13, "pressure.private_owners_degraded": 5, "pressure.private_owners_evicted": 13, "pressure.search_budget_exhaustions": 0, "pressure.searches": 28, "pressure.shared_owners_degraded": 2, "pressure.shared_owners_evicted": 2, "pressure.spill_pages": 703, "selections.private_endpoint": 0, "selections.private_long_anchor": 0, "selections.private_response_replay": 10, "selections.private_turn_closure": 0, "selections.reused_prompt_tokens": 104805, "selections.root": 12, "selections.shared_stable_prefix": 9, "state_operations.forks": 14, "state_operations.moves": 0, "state_operations.restores": 10, "state_transfers.d2d.bytes": 5, "state_transfers.d2d.count": 5, "state_transfers.d2d.seconds": 0.000382688, "state_transfers.d2h.bytes": 9, "state_transfers.d2h.count": 9, "state_transfers.d2h.seconds": 0.03214960000000001, "state_transfers.h2d.bytes": 10, "state_transfers.h2d.count": 10, "state_transfers.h2d.seconds": 0.03579872 }, "observed_occupancy_peaks": { "occupancy.device_backend_kv_pages": 0, "occupancy.device_main_kv_pages": 4407, "occupancy.device_state_slots": 8, "occupancy.host_kv_bytes": 1157554176, "occupancy.host_state_slots": 4, "occupancy.shared_active_references": 0 }, "end_occupancy": { "device_backend_kv_pages": 0, "device_main_kv_pages": 4284, "device_state_slots": 8, "host_kv_bytes": 1157554176, "host_state_slots": 4, "shared_active_references": 0 }, "nearest_rank_p50_visible_ttft_ms": 8603.36290000123, "nearest_rank_p50_e2e_ms": 8711.974600002577 }, { "pass": "revisit1", "requests": [ { "request_index": 0, "native_request_id": 32, "prompt_tokens": 62327, "completion_tokens": 265, "prefix_cache_hit_tokens": 0, "computed_prefill_tokens": 62327, "prefix_reuse_path": "root", "visible_ttft_ms": 9233.508999997866, "e2e_ms": 9324.149000000034 }, { "request_index": 1, "native_request_id": 33, "prompt_tokens": 62328, "completion_tokens": 627, "prefix_cache_hit_tokens": 0, "computed_prefill_tokens": 62328, "prefix_reuse_path": "root", "visible_ttft_ms": 10374.570799998764, "e2e_ms": 10556.770799998048 }, { "request_index": 2, "native_request_id": 34, "prompt_tokens": 62328, "completion_tokens": 343, "prefix_cache_hit_tokens": 0, "computed_prefill_tokens": 62328, "prefix_reuse_path": "root", "visible_ttft_ms": 9373.198500001308, "e2e_ms": 9482.059900001332 }, { "request_index": 3, "native_request_id": 35, "prompt_tokens": 62330, "completion_tokens": 213, "prefix_cache_hit_tokens": 0, "computed_prefill_tokens": 62330, "prefix_reuse_path": "root", "visible_ttft_ms": 8800.625100000616, "e2e_ms": 8927.579600000172 }, { "request_index": 4, "native_request_id": 36, "prompt_tokens": 62333, "completion_tokens": 517, "prefix_cache_hit_tokens": 0, "computed_prefill_tokens": 62333, "prefix_reuse_path": "root", "visible_ttft_ms": 9944.86660000257, "e2e_ms": 10035.54190000068 }, { "request_index": 5, "native_request_id": 37, "prompt_tokens": 62335, "completion_tokens": 339, "prefix_cache_hit_tokens": 0, "computed_prefill_tokens": 62335, "prefix_reuse_path": "root", "visible_ttft_ms": 8966.76590000061, "e2e_ms": 9129.797699999472 } ], "median_visible_ttft_ms": 9303.353749999587, "median_e2e_ms": 9403.104450000683, "reused_prompt_tokens": 0, "new_throughput_intervals": 13, "interval_counter_sums": { "actual_transfer_seconds": 0.034695808999999994, "backend_kv_transfers.d2d.bytes": 0, "backend_kv_transfers.d2d.pages": 0, "backend_kv_transfers.d2d.seconds": 0.0, "backend_kv_transfers.d2h.bytes": 0, "backend_kv_transfers.d2h.pages": 0, "backend_kv_transfers.d2h.seconds": 0.0, "backend_kv_transfers.h2d.bytes": 0, "backend_kv_transfers.h2d.pages": 0, "backend_kv_transfers.h2d.seconds": 0.0, "captures.aborted": 0, "captures.completed": 8, "main_kv_transfers.d2d.bytes": 9879552, "main_kv_transfers.d2d.pages": 6, "main_kv_transfers.d2d.seconds": 0.0024262719999999993, "main_kv_transfers.d2h.bytes": 1118035968, "main_kv_transfers.d2h.pages": 679, "main_kv_transfers.d2h.seconds": 0.021020992999999998, "main_kv_transfers.h2d.bytes": 0, "main_kv_transfers.h2d.pages": 0, "main_kv_transfers.h2d.seconds": 0.0, "pressure.checkpoints_dropped": 16, "pressure.historical_fork_hits": 0, "pressure.maximal_fallback_selections": 0, "pressure.partial_tail_cow_pages": 6, "pressure.private_owners_degraded": 2, "pressure.private_owners_evicted": 12, "pressure.search_budget_exhaustions": 0, "pressure.searches": 5, "pressure.shared_owners_degraded": 4, "pressure.shared_owners_evicted": 2, "pressure.spill_pages": 679, "selections.private_endpoint": 0, "selections.private_long_anchor": 0, "selections.private_response_replay": 0, "selections.private_turn_closure": 0, "selections.reused_prompt_tokens": 0, "selections.root": 6, "selections.shared_stable_prefix": 0, "state_operations.forks": 5, "state_operations.moves": 0, "state_operations.restores": 0, "state_transfers.d2d.bytes": 5, "state_transfers.d2d.count": 5, "state_transfers.d2d.seconds": 0.000417792, "state_transfers.d2h.bytes": 3, "state_transfers.d2h.count": 3, "state_transfers.d2h.seconds": 0.010830751999999999, "state_transfers.h2d.bytes": 0, "state_transfers.h2d.count": 0, "state_transfers.h2d.seconds": 0.0 }, "observed_occupancy_peaks": { "occupancy.device_backend_kv_pages": 0, "occupancy.device_main_kv_pages": 4408, "occupancy.device_state_slots": 8, "occupancy.host_kv_bytes": 1170726912, "occupancy.host_state_slots": 4, "occupancy.shared_active_references": 1 }, "end_occupancy": { "device_backend_kv_pages": 0, "device_main_kv_pages": 3239, "device_state_slots": 6, "host_kv_bytes": 1109803008, "host_state_slots": 2, "shared_active_references": 0 }, "nearest_rank_p50_visible_ttft_ms": 9233.508999997866, "nearest_rank_p50_e2e_ms": 9324.149000000034 }, { "pass": "revisit2", "requests": [ { "request_index": 0, "native_request_id": 38, "prompt_tokens": 62327, "completion_tokens": 284, "prefix_cache_hit_tokens": 62320, "computed_prefill_tokens": 7, "prefix_reuse_path": "shared_stable_prefix", "visible_ttft_ms": 1496.525800001109, "e2e_ms": 1677.395599999727 }, { "request_index": 1, "native_request_id": 39, "prompt_tokens": 62328, "completion_tokens": 265, "prefix_cache_hit_tokens": 0, "computed_prefill_tokens": 62328, "prefix_reuse_path": "root", "visible_ttft_ms": 8853.945300001214, "e2e_ms": 8980.571000000054 }, { "request_index": 2, "native_request_id": 40, "prompt_tokens": 62328, "completion_tokens": 383, "prefix_cache_hit_tokens": 0, "computed_prefill_tokens": 62328, "prefix_reuse_path": "root", "visible_ttft_ms": 9553.487699999096, "e2e_ms": 9644.11669999754 }, { "request_index": 3, "native_request_id": 41, "prompt_tokens": 62330, "completion_tokens": 1541, "prefix_cache_hit_tokens": 0, "computed_prefill_tokens": 62330, "prefix_reuse_path": "root", "visible_ttft_ms": 13643.54209999874, "e2e_ms": 13788.482299998577 }, { "request_index": 4, "native_request_id": 42, "prompt_tokens": 62333, "completion_tokens": 253, "prefix_cache_hit_tokens": 62328, "computed_prefill_tokens": 5, "prefix_reuse_path": "private_response_replay", "visible_ttft_ms": 1150.9905000020808, "e2e_ms": 1259.8532000010891 }, { "request_index": 5, "native_request_id": 43, "prompt_tokens": 62335, "completion_tokens": 185, "prefix_cache_hit_tokens": 0, "computed_prefill_tokens": 62335, "prefix_reuse_path": "root", "visible_ttft_ms": 8510.93359999868, "e2e_ms": 8637.916799998493 } ], "median_visible_ttft_ms": 8682.439449999947, "median_e2e_ms": 8809.243899999274, "reused_prompt_tokens": 124648, "new_throughput_intervals": 10, "interval_counter_sums": { "actual_transfer_seconds": 0.13131980799999998, "backend_kv_transfers.d2d.bytes": 0, "backend_kv_transfers.d2d.pages": 0, "backend_kv_transfers.d2d.seconds": 0.0, "backend_kv_transfers.d2h.bytes": 0, "backend_kv_transfers.d2h.pages": 0, "backend_kv_transfers.d2h.seconds": 0.0, "backend_kv_transfers.h2d.bytes": 0, "backend_kv_transfers.h2d.pages": 0, "backend_kv_transfers.h2d.seconds": 0.0, "captures.aborted": 0, "captures.completed": 7, "main_kv_transfers.d2d.bytes": 8232960, "main_kv_transfers.d2d.pages": 5, "main_kv_transfers.d2d.seconds": 0.002419743999999999, "main_kv_transfers.d2h.bytes": 2058240000, "main_kv_transfers.d2h.pages": 1250, "main_kv_transfers.d2h.seconds": 0.036301631, "main_kv_transfers.h2d.bytes": 1017593856, "main_kv_transfers.h2d.pages": 618, "main_kv_transfers.h2d.seconds": 0.018998432000000003, "pressure.checkpoints_dropped": 7, "pressure.historical_fork_hits": 1, "pressure.maximal_fallback_selections": 0, "pressure.partial_tail_cow_pages": 5, "pressure.private_owners_degraded": 0, "pressure.private_owners_evicted": 2, "pressure.search_budget_exhaustions": 0, "pressure.searches": 5, "pressure.shared_owners_degraded": 4, "pressure.shared_owners_evicted": 3, "pressure.spill_pages": 1250, "selections.private_endpoint": 0, "selections.private_long_anchor": 0, "selections.private_response_replay": 1, "selections.private_turn_closure": 0, "selections.reused_prompt_tokens": 124648, "selections.root": 4, "selections.shared_stable_prefix": 1, "state_operations.forks": 4, "state_operations.moves": 0, "state_operations.restores": 1, "state_transfers.d2d.bytes": 3, "state_transfers.d2d.count": 3, "state_transfers.d2d.seconds": 0.0002876799999999999, "state_transfers.d2h.bytes": 5, "state_transfers.d2h.count": 5, "state_transfers.d2h.seconds": 0.01931456000000001, "state_transfers.h2d.bytes": 1, "state_transfers.h2d.count": 1, "state_transfers.h2d.seconds": 0.05399776100000001 }, "observed_occupancy_peaks": { "occupancy.device_backend_kv_pages": 0, "occupancy.device_main_kv_pages": 4408, "occupancy.device_state_slots": 8, "occupancy.host_kv_bytes": 2056593408, "occupancy.host_state_slots": 4, "occupancy.shared_active_references": 1 }, "end_occupancy": { "device_backend_kv_pages": 0, "device_main_kv_pages": 4283, "device_state_slots": 8, "host_kv_bytes": 2056593408, "host_state_slots": 4, "shared_active_references": 0 }, "nearest_rank_p50_visible_ttft_ms": 8510.93359999868, "nearest_rank_p50_e2e_ms": 8637.916799998493 } ], "revisit_population": { "requests": 12, "median_visible_ttft_ms": 9100.137449999238, "median_e2e_ms": 9226.973349999753, "reused_prompt_tokens": 124648, "nearest_rank_p50_visible_ttft_ms": 8966.76590000061, "nearest_rank_p50_e2e_ms": 9129.797699999472 }, "host_memory": { "current_bytes": 24800956416, "effective_limit_bytes": 42949672960, "effective_swap_limit_bytes": 0, "events": { "high": 0, "low": 0, "max": 0, "oom": 0, "oom_group_kill": 0, "oom_kill": 0 }, "exit_code": 0, "limit_bytes": 42949672960, "oom_killed": false, "peak_bytes": 24801497088, "ram_plus_swap_limit_bytes": 42949672960 }, "limits": [ "First pass interval sums include preflight intervals; later passes sum only newly captured native interval records. Occupancy is a sampled gauge, not an exact continuous peak.", "Throughput interval attribution assumes root-reported quiet>=6s capture boundaries; native IDs prove no intervening inference.", "All normal records are private until sanitized; native errors/startup can contain identifying free text.", "Observed state_transfers.*.bytes numerically equal operation counts in control logs. Preserve raw field names but do not use these as physical state-byte volumes without producer-unit verification; main_kv_transfers bytes/pages are separately reported." ], "sources": { "swift-cache-control-cold.json": "974d3451181377b04c76f27224238cc51845b9157c78aebe0b48be3113a04267", "swift-cache-control-revisit1.json": "dd2fdffd7ba1777b9394fc6cef65b132697bb3daa1c64a7b20a3ff58af9e6317", "swift-cache-control-revisit2.json": "206702b7ba4d9d069b7b6e262796569808602ccb0168e95a3d927fc6f3a76da9", "swift-cache-control-cold-native.json": "92112f268ee6ec286c604431b40b87599da4889685d00d91ab15f0dacfe4b16c", "swift-cache-control-revisit1-native.json": "56860f733070da83341baade2b9d19a410034b53ffd9da512826127b2d0431da", "swift-cache-control-revisit2-native.json": "1f98cde65cc3bf3ecc8ee3661c54afe6f3d83e472bfffaa12289bf04ce6d3875", "swift-cache-control-preflight.json": "4bef37cae1a171ec1719c1ae16b228172db65b03ff008cc6aa57c4f6d4baed6e", "swift-cache-control-final-status.json": "1692c7b7b85ce11806a688aa7ba7e9ea8911d187be0e9619e526ad3a2eca1261" }, "median_convention": "Statistical median averages the two middle observations for even N; native nearest-rank p50 is also retained. Gate evaluation must report both, never choose the favorable one." }