PERFORMANCE BENCHMARK OF CEPH FILE SYSTEM 17.2.5

• Main Author: Rizaldi Rahardjo, Tanur
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/78108
• Institution: Institut Teknologi Bandung

Ceph are storage platform that is used by clientssuch as Open Stack and CERN for their storage backend. Ceph provides many storage interfaces that can be used by clients for various use cases, This paper will focus on the file system interface Ceph File System or CephFS. Previous performance benchmark research test CephFS 15 metadata performance. At the time this research was conducted, Ceph already released version 17 that contain various improvements. A few notable improvements in Ceph 17 are the total migration for RADOS database backend from LevelDB to RocksDB and Ceph 16 also introduce a new metadata format for CephFS that might affect CephFS performance. This research will reinvestigate CephFS metadata service (MDS) low performance based on previous research result. Ceph 17 various improvement might potentially improve CephFS performance by a significant margin compared with Ceph 15. This research will benchmark the metadata performance of CephFS 17.2.5 (Quincy) and 15.2.17 (Octopus). This research will include a throughput benchmark for CephFS using the same cluster configuration with metadata benchmark. By having both metadata and throughput benchmark, this research will give more information on whether CephFS 17 is suitable for specific high performance computing use case. As a bonus, this research will include an exploration of CephFS features, notably multiple active MDS and LazyIO. Experiment results show that the performance of Ceph 17.2.5 is improved dramatically at certain tasks, but slightly lower at a few other tasks.