INTELLIGENT ESTIMATION AND PREDICTIVE MINIMIZATION OF DELAYS IN SERVERLESS PLATFORMS BASED ON HYBRID NEURAL NETWORK ARCHITECTURE CNN-LSTM

Abstract

The article investigates the problem of cold start in Serverless platforms as one of the main reasons for the
deterioration of service quality indicators under conditions of uneven and dynamic load. This problem is particularly
difficult for Function-as-a-Service environments, where the stochastic nature of function calls and the presence of periods
of inactivity lead to significant delays in the initialization of the computing context. Traditional heuristic resource
management policies, in particular fixed timeouts or reactive scaling, often prove ineffective under conditions of sharp
load surges. An approach to predictive resource management of Serverless platforms based on a hybrid CNN–LSTM
neural network architecture designed to predict the intensity of function calls is proposed. Convolutional neural networks
are used to highlight local patterns and short-term peaks in the load time series, while recurrent LSTM blocks provide
modeling of long-term time dependencies. The obtained forecasts are used to implement a predictive pre-warming
mechanism, which allows to activate computing resources in advance and minimize cold start delays. Experimental
research was conducted on real Azure Functions execution routes using scenarios of uneven load and periods of inactivity.
The experimental results showed that the proposed CNN–LSTM model provides a 29.3% reduction in the mean square
prediction error compared to the standard LSTM and allows to reduce the number of cold starts by an average of 32.4%.
The obtained results confirm the feasibility of using hybrid neural network approaches to improve the efficiency and
quality of service in modern Serverless systems.
Keywords: serverless computing, Cold Start, CNN-LSTM, load forecasting, Function-as-a-Service, pre-warming,
deep learning.

References
1. Box, G. E. P., Jenkins, G. M., Reinsel, G. C., & Ljung, G. M. (2015). Time Series Analysis: Forecasting and
Control (5th ed.). John Wiley & Sons. URL: https://books.google.com.ua/books?id=De9xBgAAQBAJ.
2. Jonas, E., et al. (2019). Cloud Programming Simplified: A Berkeley View on Serverless Computing. arXiv
preprint arXiv:1902.03383. URL: https://arxiv.org/abs/1902.03383.
3. Baldini, I., et al. (2017). Serverless Computing: Current Trends and Open Problems. Research Advances in
Cloud Computing, 1–20. URL: https://arxiv.org/abs/1706.03178.
4. Hochreiter, S., & Schmidhuber, J. (1997). Long Short-Term Memory. Neural Computation, 9(8), 1735–1780.
DOI: 10.1162/neco.1997.9.8.1735.
5. Lin, C., & Khazaei, H. (2021). Modeling and Optimization of Performance and Cost of Serverless
Applications. IEEE Transactions on Parallel and Distributed Systems, 32(3), 655–669. DOI:
10.1109/TPDS.2020.3028841.
6. Osypanka, P., & Nawrocki, P. (2022). Resource Usage Cost Optimization in Cloud Computing Using Machine
Learning. IEEE Transactions on Cloud Computing, 10(3), 2079–2089. DOI: 10.1109/TCC.2020.2987821.
7. O'Shea, K., & Nash, R. (2015). An Introduction to Convolutional Neural Networks. arXiv preprint
arXiv:1511.08458. URL: https://arxiv.org/abs/1511.08458.
8. Goodfellow, I., Bengio, Y., & Courville, A. (2016). Deep Learning. MIT Press. URL:
http://www.deeplearningbook.org.
9. Wang, L., Li, M., Zhang, Y., Ristenpart, T., & Swift, M. (2018). Peeking Behind the Curtains of Serverless
Platforms. 2018 USENIX Annual Technical Conference (USENIX ATC 18), 133–146. URL: https://www.usenix.
org/conference/atc18/presentation/wang-liang.
10. Karim, F., Majumdar, S., Darabi, H., & Chen, S. (2018). LSTM Fully Convolutional Networks for Time Series
Classification. IEEE Access, 6, 1662–1669. DOI: 10.1109/ACCESS.2017.2779939.
11. Shahrad, M., et al. (2020). Serverless in the Wild: Characterizing and Optimizing the Serverless Workload at
a Large Cloud Provider. 2020 USENIX Annual Technical Conference (USENIX ATC 20), 205–218. URL:
https://www.usenix.org/conference/atc20/presentation/shahrad.