Activeness-Based Data Retention Recommender for HPC Facilities

Student: Wei Zhang (Texas Tech University)
Supervisor: Yong Chen (Texas Tech University)

Abstract: The storage system of many high-performance computing (HPC) facilities faces an increasingly challenging goal of meeting unlimited data growth with limited storage capacity growth. The data retention policy plays a vital role in addressing such challenge. However, most existing data retention policies are designed based on and biased towards the temporal properties of files rather than the activities of users. Such biased data retention policies may cause unnecessary data interruption or even data loss to users. In this study, we propose a data retention action recommender system (DataRecommender) that generates user-centric data retention recommendations based on a holistic view of the user activeness. Our evaluation on the recommendation results shows that, as compared to the existing data retention policies, adopting DataRecommender’s data retention recommendation can significantly avoid undesired data loss for active users while maintaining the same ability of cleaning unnecessary files as compared to existing LRU-based data retention policies.

ACM-SRC Semi-Finalist: no

Poster: PDF
Poster Summary: PDF


Back to Poster Archive Listing