DSTFView: AI Workload Forecasting for Cloud-Edge Systems
A new framework called DSTFView has been introduced by researchers for forecasting workloads in collaborative cloud-edge settings. This dual-input spatio-temporal-frequency model effectively captures both proximity and periodic dependencies, while also extracting features from spatial, temporal, and frequency domains. Additionally, it utilizes an adaptive fusion mechanism to detect sudden changes. Testing on CPU and TP datasets indicates that DSTFView surpasses key baseline models across various forecasting timeframes and evaluation criteria. This research tackles the issue of achieving an effective balance between modeling multidimensional features and ensuring forecasting efficiency in cloud-edge systems that cater to latency-sensitive, highly concurrent, and reliability-critical applications.
Key facts
- DSTFView is a dual-input spatio-temporal-frequency multi-view workload forecasting framework.
- It models closeness and period dependencies.
- It extracts spatial, temporal, and frequency-domain dependencies.
- It uses an adaptive fusion mechanism to adjust view contributions for abrupt changes.
- Experiments were conducted on CPU and TP datasets.
- DSTFView outperforms representative baselines across multiple forecasting horizons.
- The framework targets collaborative cloud-edge environments.
- Edge platforms increasingly support latency-sensitive, highly concurrent, and reliability-critical applications.
Entities
Institutions
- arXiv