Research map / Graph learning / Knowledge and temporal graphs
SMART: A Surrogate Model for Predicting Application Runtime in Dragonfly Systems
Xin Wang, Pietro Lodi Rizzini, Sourav Medya, Zhiling Lan
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add online tuning every 8 iterations (F_t=8) | SMART · D1 MILC contiguous | MAPE (%): 3.98 -> 3.70 | Improved |
| increase LLM input window T_LLM from 2 to 8 | SMART · D1 MILC contiguous | MAPE (%): 3.70 -> 3.19 | Improved |
| add the Time-LLM component (versus GNN-only) (ablation) | SMART · D1 LAMMPS contiguous | MAPE (%): 2.05 -> 1.78 | Improved |
| add the GNN encoder (versus LLM-only) (ablation) | SMART · D1 MILC contiguous | MAPE (%): 4.02 -> 3.19 | Improved |
| use Time-LLM instead of Autoformer (ablation) | SMART · D2 MILC contiguous | MAPE (%): 4.22 -> 3.91 | Improved |
| use Time-LLM instead of DLinear (ablation) | SMART · D2 NN contiguous | MAPE (%): 4.11 -> 3.07 | Improved |
| use Time-LLM instead of PatchTST (ablation) | SMART · D2 NN contiguous | MAPE (%): 3.15 -> 3.07 | Improved |
| use Time-LLM instead of Autoformer (ablation) | SMART · D2 LAMMPS contiguous | MAPE (%): 1.93 -> 1.84 | Hurt |
| increase GCN layers beyond two (ablation) | SMART · D1/D2 | MAPE (%) | No clear effect |
| increase LLM patch length (ablation) | SMART · D1/D2 | MAPE (%) | Hurt |
| vary LLM hidden layers (ablation) | SMART · D1/D2 | MAPE (%) | No clear effect |
| use SMART instead of the best baseline | SMART · D1 | MAPE (%): 6.09 -> 3.19 | Improved |
Closest papers by meaning
- EvoMesh: Adaptive Physical Simulation with Hierarchical Graph Evolutions (ICML 2025 · similarity 0.82)
- Surprisingly Strong Performance Prediction with Neural Graph Features (ICML 2024 · similarity 0.82)
- Tackling the Data-Parallel Load Balancing Bottleneck in LLM Serving: Practical Online Routing at Scale (NeurIPS 2026 · similarity 0.81)
- FastPERT: Towards Fast Microservice Application Latency Prediction via Structural Inductive Bias over PERT Networks (AAAI 2025 · similarity 0.80)
- RoutingBench: Can Agentic Routing Analysis Scale to Production Datacenter Networks? (NeurIPS 2026 · similarity 0.80)
- DCcluster-Opt: Benchmarking Dynamic Multi-Objective Optimization for Geo-Distributed Data Center Workloads (NeurIPS 2025 · similarity 0.80)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
