Effective selection of completely fair scheduler algorithm in RAID kernel for improved I/O performance using machine learning

The widespread adoption of cloud-based RAID (Re- dundant Array of Independent Disks) technologies is a result of the rapid growth in data and the increasing need for high- performance storage. This paper aims to investigate the working model of a completely fair scheduler algorithm in cloud-based RAID systems that uses a machine learning-based approach to select the most suitable RAID level, optimizing I/O performance. We provide an overview of several proposed techniques such as file classification and separation, deployment of files on multiple SSDs, machine learning for RAID performance measurement and I/O isolation scheme. //To prevent kernel threads from blocking user-thread parallelism drop-offs, scheduling must be implemented carefully. When user-threading parallelism drops off, idle kernel-level threads should be put to sleep at appropriate times to avoid wasting CPU resources. Furthermore, it is crucial that the scheduling system provides fairness while preventing any single user thread from monopolizing a kernel thread; otherwise, other user threads may experience short/long term starvation or kernel threads can deadlock waiting for events to occur on busy kernel threads. Our model aims to improve RAID performance by identifying the most suitable RAID level for different workloads. We developed a RAID prediction model based on analysis using various machine learning algorithms including Random Forest, XGBoost, Decision Tree and KNN. In conclusion, our research proposes an innovative solution that utilizes machine learning algorithms as part of completely fair scheduler algorithms which can optimize I/O performance by selecting optimal RAIDs suited towards specific workloads resulting in improved overall system performance.

Перевод пока недоступен