Unistore A Unified Storage Architecture for Cloud Computing

Unistore: A Unified Storage Architecture for Cloud Computing Wei Xie, Jiang Zhou, and Yong Chen Unistore: review of project plan Ø Based on sheepdog distributed store for virtual machine Ø Optimization for heterogeneous storage (SSDs and HDDs) Ø Optimization for heterogeneous workload Sheepdog: object storage Benchmark Ø Planned testing tools Ø fio, dd for generating synthetic workload Ø Real workload benchmark Ø iostat Ø Comparison with other product Ø Gluster. FS Ø Ceph Ø Object storage Workload characterization Ø Hot/cold data detection and separation Sheepdog: gateway Planned Schedule Ø Year: 2015 Ø Q 1: investigation and survey about Unistore Ø Q 2: characterization component development of Unistore Ø Q 3: metadata management of Unistore Ø Q 4: data distribution management of Unistore Ø Year: 2016 Ø Q 1: VM image store and loading Ø Q 2: advanced functions of Unistore Ø Q 3: performance optimization of Unistore Ø Q 4: module integration and system benchmarking finished on-going Ø Multiple bloom filter [1] Ø Temporal locality [2] Ø Responsible for where to store objects, or data placement Ø Consistent hashing Ø Add/remove node not significantly change mapping Ø I/O load balance Ø How to make consistent hashing support heterogonous device? Ø Two hash rings for HDD and SSD, respectively Write_ops Write_to_HDD Ø I/O size, write/read ratio, inter-arrival time, queue depth, latency, and IOPS Ø Online vs. . offline workload tracing and characterization Initial Characterization Result Ø Plan to implement online hot/cold data detection Ø Hot data store on SSD, cold on HDD Ø Initial result collected from OLTP I/O trace. Write_to_SSD to-be-done Sheepdog: component Ø Cluster manager Ø QEMU block Ø Driver Ø Object storage Ø Gateway Ø Object manager Deployment Ø Environment Ø 3 Cent. OS 6. 5 virtual machines on i. Mac workstation Ø Sheepdog built on the 3 virtual machine and form a cluster Ø Use corosync to manage the cluster (can switch to zookeeper if necessary) Ø Will migrate to a real Linux cluster later (for testing) Acknowledgements We are grateful to the Nimboxx and the Cloud and Autonomic Computing site at Texas Tech University for the valuable support for this project.