SIRIS: Can LLMs Tell When Simulation Errors Matter in AI-Driven Systems Research
Organizations: KAIST · Microsoft Research Asia
Abstract
LLM agents increasingly drive systems research and rely on simulators to evaluate candidate designs at scale. Simulators are imperfect by design: their modeling limitations can distort predictions yet leave the recommended design correct, or change which design should be selected. We introduce Siris to test whether LLMs can tell these cases apart, judging from their systems understanding when a limitation invalidates the simulator's recommendation. By perturbing a simulator, Siris constructs workload pairs with the same candidates, limitation, and simulator recommendation. The perturbation changes predictions on both workloads, yet a reference simulator supports the recommendation on only one. Evaluating thirteen frontier LLMs on LLM serving design problems, we find that identifying a limitation does not ensure a correct choice, and even the strongest models sometimes abandon valid recommendations. Self-correction through a review turn helps these models restore valid choices but introduces new errors in several others. These findings motivate evaluating systems understanding through design decisions under imperfect simulators.
Figures & tables
| Simulator | Omitted or approximated behavior | Simulator | Omitted or approximated behavior |
|---|---|---|---|
| APEX | Tokenization and position embeddings | FlexFlow | Runtime overhead, full-bandwidth transfers |
| Vidur | Per-request context lengths as total KV reads | TopoOpt | Packet behavior during search |
| LLMServingSim | Per-block compilation, GPU kernel tuning | SimGrid | TCP slow start, packet-level behavior |
| ASTRA-sim 2.0 | Congestion, in the analytical network model | FireSim | Storage timing, as a functional block device |
| GenZ | Microarchitecture and expert-load imbalance | BookSim2 | Full-system traffic, in trace-replay mode |
Appendix figures & tables19 assets
Supplementary material from the paper’s appendix.
Appendix
| Conversations | ||||
| Problem kind | Pairs | Problems | Per cond. | Total |
| Request routing | 12 | 24 | 936 | 2,808 |
| Prefix cap | 6 | 12 | 468 | 1,404 |
| Reuse granularity | 6 | 12 | 468 | 1,404 |
| Handoff | 6 | 12 | 468 | 1,404 |
| Shared NIC | 6 | 12 | 468 | 1,404 |
| Problem kind | Perturbed computation | Change form |
| Request routing : which replica serves each request, among load-only, SGLang-inspired, Preble, and DualMap | ||
| Prefix cap | Prefix reuse is credited only up to 512 tokens | Value or constant |
| Reuse granularity | Prefix reuse is credited only in 256-token units | Value or constant |
| Prefill-to-decode handoff : where decode runs after prefill, among stay on source and three remote placements | ||
| Shared NIC | Transfers use a dedicated link per GPU pair instead of a shared NIC | Omission |
| Local scheduling : the prefill chunk cap per step, among 16, 24, and 32 tokens and no cap | ||
| Upheld | Overturned | |||||||
|---|---|---|---|---|---|---|---|---|
| Model | Init. | Fin. | Init. | Fin. | ||||
| GPT-6 Astra | 67 | 14 | 1 | 80 | 86 | 0 | 0 | 86 |
| GPT-6.1 Sol | 57 | 13 | 1 | 69 | 80 | 1 | 0 | 81 |
| GPT-6 Sol | 81 | 3 | 2 | 82 | 14 | 18 | 0 | 32 |
| GPT-5.6 Sol | 86 | 0 | 10 | 76 | 3 | 31 | 0 | 34 |
| GPT-5.6 Terra | 87 | 0 | 15 | 72 | 2 | 17 | 0 | 19 |
| No-sim | Ref. | Perturbed | ||||
|---|---|---|---|---|---|---|
| Model | Up. | Ov. | ||||
| GPT-6 Astra | 34 | 100 | 92 | 99 | 95 | |
| GPT-6.1 Sol | 37 | 100 | 79 | 93 | 86 | |
| GPT-6 Sol | 37 | 99 | 94 | 37 | 66 | |
| GPT-5.6 Sol | 30 | 100 | 87 | 39 | 63 | |
| GPT-5.6 Terra | 36 | 100 | 83 | 22 | 52 | |
| Evidence | Up. | Ov. | Pair | Follow | ||
|---|---|---|---|---|---|---|
| GPT-5.6 Sol | ||||||
| None | 39 | 22 | 30 | 10 | – | |
| Code | 70 | 48 | 59 | 37 | 31 | |
| Code + records | 83 | 10 | 47 | 8 | 69 | |
| Code + predictions | 83 | 28 | 55 | 22 | 59 | |
| Code + both | 87 | 39 | 63 | 34 | – | |
| Any turn | First turn | |||
|---|---|---|---|---|
| Model | Pert. | Ref. | Pert. | Ref. |
| GPT-6 Astra | 100 | 99 | 100 | 97 |
| GPT-6.1 Sol | 100 | 100 | 100 | 97 |
| GPT-6 Sol | 74 | 76 | 34 | 38 |
| GPT-5.6 Sol | 80 | 80 | 4 | 0 |
| GPT-5.6 Terra | 82 | 74 | 13 | 6 |
| Model | Prompt | Reach | Ident. | Up. | Ov. | Pair |
|---|---|---|---|---|---|---|
| GPT-6.1 Sol | Standard | 174 | 171 | 69 | 81 | 65 |
| Instructed | 174 | 171 | 67 | 82 | 65 | |
| GPT-5.6 Sol | Standard | 7 | 112 | 76 | 34 | 30 |
| Instructed | 173 | 166 | 60 | 59 | 40 | |
| GPT-5.6 Luna | Standard | 4 | 82 | 86 | 16 | 16 |
| Instructed | 162 | 123 | 78 | 21 | 18 |
| Perturbed | Ref. | No-sim | ||||
|---|---|---|---|---|---|---|
| Model | Up. | Ov. | All% | Pair | Pair | Pair |
| GPT-6 Astra | 80 | 86 | 95.4 | 79 | 87 | 6 |
| GPT-6.1 Sol | 69 | 81 | 86.2 | 65 | 87 | 8 |
| GPT-6 Sol | 82 | 32 | 65.5 | 32 | 86 | 14 |
| GPT-5.6 Sol | 76 | 34 | 63.2 | 30 | 87 | 9 |
| GPT-5.6 Terra | 72 | 19 | 52.3 | 14 | 87 | 10 |
| Identified | Not | |||||
|---|---|---|---|---|---|---|
| Model | All | Corr. | Up. | Pres. | Third | ident. |
| GPT-6 Astra | 174 | 166 | 7 | 0 | 1 | 0 |
| GPT-6.1 Sol | 171 | 150 | 18 | 0 | 3 | 3 |
| GPT-6 Sol | 95 | 72 | 5 | 15 | 3 | 79 |
| GPT-5.6 Sol | 112 | 79 | 11 | 21 | 1 | 62 |
| GPT-5.6 Terra | 101 | 59 | 15 | 23 | 4 | 73 |
| Batch 1 | Batch 2 | |||||
|---|---|---|---|---|---|---|
| Condition | Up. | Ov. | Pair | Up. | Ov. | Pair |
| Predictions only | 55 | 7 | 2 | 49 | 6 | 1 |
| Full, standard | 53 | 36 | 31 | 55 | 35 | 30 |
| Full, comparison | 59 | 38 | 37 | 60 | 37 | 37 |
| Provided to the agent | Held by the evaluator |
|---|---|
| System specification | Perturbation type and location |
| Candidate code, workload, target metric | Reference implementation and source difference |
| Presented simulator code, cost parameters, configuration | Reference execution, metrics, and winner |
| Presented predictions and execution records | Pair partner’s packet and reference answer |
| Read-only search and reading tools | Upheld or overturned label and pair membership |
| Design | Behavior | Upheld (s) | Overturned (s) |
|---|---|---|---|
| Load-only | Sends each request to the replica with the fewest waiting and running requests, then the fewest queued tokens, and never looks at the caches. | 1.050 / 0.703 | 0.326 / 0.230 |
| SGLang-inspired | Sends each request to the replica whose cache holds the longest prefix of its prompt, breaking ties by the fewest requests and then the fewest tokens. | 1.597 / 1.233 | 0.552 / 0.180 |
| Preble | Scores each replica by its recent load, the prompt tokens it would prefill, and the cost of evicting cached prefixes weighted by recent use, and when the best reuse covers more than half of the prompt, only replicas with that reuse compete. | 3.445 / 2.737 | 0.763 / 0.660 |
| DualMap | Hashes the prompt prefix to two candidate replicas and takes the one with more cached reuse if its predicted prefill time meets a 0.6 s TTFT target, otherwise the one with the lower predicted time. | 1.800 / 1.327 | 0.718 / 0.209 |
| Design | Behavior | Upheld (s) | Overturned (s) |
|---|---|---|---|
| Stay on source | Decodes on the prefill replica and never moves the KV cache. | 3.465 / 3.465 | 1.816 / 1.816 |
| Next instance | Moves the decode to the next replica above the prefill replica. | 3.771 / 3.771 | 2.424 / 2.424 |
| Least-active remote | Moves the decode to the remote replica with the fewest active calls, not counting transfers still in flight. | 3.769 / 3.769 | 2.425 / 2.425 |
| Least-outstanding | Counts active calls plus handoffs sent but not yet arrived, and moves the decode to the remote replica with the fewest. | 2.826 / 3.030 | 1.649 / 2.056 |
LLM Decision-Making
LLM decision-making studies how language models choose, act, seek information, and allocate resources under goals, uncertainty, constraints, and social pressures.
See all 275 papers