Docker platform aging: a systematic performance evaluation and prediction of resource consumption

Docker platform aging: a systematic performance evaluation and prediction of resource consumption

---

연구 분야: Software Development

---

---

학회: The Journal of Supercomputing

---

초록

Software aging is a complex phenomenon that manifests itself in software-based applications and platforms as steady performance and resource deterioration. Previous research has made substantial progress in modeling, assessment, and experimental studies of software aging phenomena in software systems such as virtualized server systems, but has not taken into account software platforms in a platform as a service (PaaS), such as Docker. The Docker platform, which is used to run OS-level virtualization-based containers, is prone to software aging in all cases, necessitating more research. This paper presents a thorough investigation into software aging and rejuvenation, as well as behavior prediction on Docker platform. For 30 days, a collection of software aging experiments was executed continuously to explore the accumulation and variation of aging-related problems on different hardware configurations. Furthermore, in software aging and software rejuvenation studies, the Stress-Wait-Rejuvenation (SWARE) approach is used to detect aging signs and examine the efficiency of software rejuvenation strategies in a single trial. We can integrate the software platform’s resource usage under the workload in this way. Even after the stressful workload is stopped, the resource consumption remains significant, according to the findings. Experiments show that the Docker platform has had serious aging issues since its early phases of implementation. However, rejuvenation approaches, such as just rebooting the operating system, can drastically cut down on resource utilization. For predicting aging phenomena in the chosen Docker platform and comparing aging prediction models, the experimental data are gathered and supplied to different data prediction models based on time series and a long short-term memory (LSTM) machine-learning algorithm. The prediction results show that the LSTM model beats the other examined models in terms of aging forecast accuracy. The findings of this study can aid in the management and maintenance of Docker-based PaaS systems by taking into account the effects of software aging and using appropriate software rejuvenation strategies in Docker platforms.

---