Application-oriented cloud workload prediction: A survey and new perspectives

With the rapid development of cloud computing, more and more applications have migrated or will migrate to cloud platforms. Cloud platforms provide applications with powerful computing capabilities, flexible resource allocation, and high scalability, ensuring that applications can achieve better performance, cost, and energy efficiency. In addition, cloud computing brings other advantages to applications, such as global deployment, high availability, and robust data processing capabilities, enabling applications to flexibly meet user requirements, deliver faster and more stable services, and provide a seamless experience for users worldwide. However, due to dynamic cloud environments, diverse user requests and services, and elastic cloud resource provisioning, the resource management for cloud applications faces significant challenges, and applications often suffer from long queue time, erratic performance, resource contention and idleness, and high energy consumption.

 

To address the above problems, the MAPE cycle, a proactive application intelligence O&M framework, has been proposed and promoted. It consists of four steps: monitoring, analysis, planning, and execution, where the analysis step is to predict the future workloads of applications and guide the proper resource management decisions in the planning step to avoid QoS degradation, cost and energy inefficiencies, etc. Therefore, workload prediction is a critical step in intelligent O&M and is essential for execution quality assurance throughout the application lifecycle.

 

A team led by Prof. Zhijun Ding at Tongji University, China, recently reviewed the latest cloud workload prediction papers and related key technologies. This work classifies existing cloud workload prediction work from the new perspective of "application-oriented," deeply analyses the variability and heterogeneity of application workloads, and systematically explains how existing work can guide workload prediction by using application evolution characteristics and how to guide proactive long- and short-term resource management practices based on workload prediction results. Based on this, it summarises the technical challenges and future research directions that have yet to be addressed by existing research.

 

The team published their review in Tsinghua Science and Technology on 11 September 2024.

 

Specifically:

(1) It provides an overview of the basic features associated with workload prediction, including prediction goals, modeling techniques, evaluation metrics, and datasets.

(2) It analyses two characteristics of cloud applications, including variability and heterogeneity, and how application-specific features affect their workload variations. And it classifies recently published work on workload prediction based on the features of cloud applications as well as research ideas, summarises research motivations, main contributions and core ideas.

(3) It classifies work on proactive cloud application resource management based on workload prediction, including long-term management, such as proactive capacity planning, application deployment, and dynamic migration, and short-term management, such as proactive request scheduling, resource allocation, and elastic scaling.

(4) It reveals outstanding technical challenges and potential development opportunities in current research on workload prediction, including the following aspects: large-scale application, serverless application, multi-topology awareness, large prediction model, model interpretability, and model unreliability.

 

The authors include Binbin Feng and Zhijun Ding from the Key Laboratory of Embedded System and Service Computing, Ministry of Education, and the Department of Computer Science and Technology at Tongji University in Shanghai, China.

 

This work was supported by the National Natural Science Foundation of China (62372330).

 

---

About the Authors

Zhijun Ding received the PhD degree in computer application technology from Tongji University, Shanghai, China, in 2007. Currently, he is a professor with the Department of Computer Science and Technology, Tongji University, Shanghai, China. His research interests include formal methods, Petri nets, services computing, and workflow. He has published more than 100 papers in domestic and international academic journals and conference proceedings.

 

Binbin Feng is a doctoral student in the Department of Computer Science and Technology at Tongji University. His research direction is cloud workload online prediction method and its application. Until now, he has published academic papers at TPDS, TSC, TSE journals and WWW, ICWS and IEEE CLOUD conferences.

 

[1] B. Feng and Z. Ding, "Application-Oriented Cloud Workload Prediction: A Survey and New Perspectives," in Tsinghua Science and Technology, vol. 30, no. 1, pp. 34-54, February 2025, doi: 10.26599/TST.2024.9010024.

 

---

About Tsinghua Science and Technology

Tsinghua Science and Technology is sponsored by Tsinghua University and published bimonthly, 2023 Impact Factor of 5.2, ranking in Q1 in the "Computer Science, Software Engineering", "Computer Science, Information System", and "Engineering, Electrical & Electronic" areas in SCIE, according to JCR 2023. This journal aims at presenting the achievements in computer science, electronic engineering, and other IT fields. This journal has been indexed by SCIE, EI, Scopus, etc. Contributions all over the world are welcome.

About SciOpen 

SciOpen is an open access resource of scientific and technical content published by Tsinghua University Press and its publishing partners. SciOpen provides end-to-end services across manuscript submission, peer review, content hosting, analytics, identity management, and expert advice to ensure each journal’s development. By digitalizing the publishing process, SciOpen widens the reach, deepens the impact, and accelerates the exchange of ideas.

---

---