TY - GEN
T1 - HCFS2
T2 - 12th IEEE International Symposium on Service-Oriented System Engineering, SOSE 2018 and 9th International Workshop on Joint Cloud Computing, JCC 2018
AU - Sun, Jie
AU - Hu, Chunming
AU - Wo, Tianyu
AU - Du, Lele
AU - Yang, Song
N1 - Publisher Copyright:
© 2018 IEEE.
PY - 2018/5/14
Y1 - 2018/5/14
N2 - In the hybrid cloud, multiple private and public clouds usually communicate through Wide Area Networks (WAN), suffering from high latency and low bandwidth for inter-cloud data transmission. While existing DFSs are widely used in a single cloud, they may bring significant I/O delay in the hybrid cloud where, based on our investigation, applications are sensitive to latency but barely rely on strictly consistent storage. However, existing DFSs are mainly designed with strong consistency semantics. To address this problem, we implement HCFS2 (Hybrid Cloud File Storage Service). We reuse some components of MooseFS while weakening its consistency semantics. HCFS2 holds three main features: (1) It generates file update digests through intercepting and parsing client I/O operations in userspace, and leverages gossip protocol to distribute file update digests among geographically distributed servers. (2) It maintains weak consistency among storage servers in the hybrid cloud, and it uses a two-level consistency setup to ensure local consistency and global consistency, respectively. (3) It maintains three Log queues to simplify the Log management and uses different task queues to parallelize different processing stages for synchronization, thus improving its concurrent operation performance. Experiment results show that HCFS2 achieves good performance in the hybrid cloud.
AB - In the hybrid cloud, multiple private and public clouds usually communicate through Wide Area Networks (WAN), suffering from high latency and low bandwidth for inter-cloud data transmission. While existing DFSs are widely used in a single cloud, they may bring significant I/O delay in the hybrid cloud where, based on our investigation, applications are sensitive to latency but barely rely on strictly consistent storage. However, existing DFSs are mainly designed with strong consistency semantics. To address this problem, we implement HCFS2 (Hybrid Cloud File Storage Service). We reuse some components of MooseFS while weakening its consistency semantics. HCFS2 holds three main features: (1) It generates file update digests through intercepting and parsing client I/O operations in userspace, and leverages gossip protocol to distribute file update digests among geographically distributed servers. (2) It maintains weak consistency among storage servers in the hybrid cloud, and it uses a two-level consistency setup to ensure local consistency and global consistency, respectively. (3) It maintains three Log queues to simplify the Log management and uses different task queues to parallelize different processing stages for synchronization, thus improving its concurrent operation performance. Experiment results show that HCFS2 achieves good performance in the hybrid cloud.
KW - Distributed File System
KW - File Synchronization
KW - Gossip Protocol
KW - Hybrid Cloud
KW - Weak Consistency
UR - https://www.scopus.com/pages/publications/85048279874
U2 - 10.1109/SOSE.2018.00038
DO - 10.1109/SOSE.2018.00038
M3 - 会议稿件
AN - SCOPUS:85048279874
T3 - Proceedings - 12th IEEE International Symposium on Service-Oriented System Engineering, SOSE 2018 and 9th International Workshop on Joint Cloud Computing, JCC 2018
SP - 228
EP - 233
BT - Proceedings - 12th IEEE International Symposium on Service-Oriented System Engineering, SOSE 2018 and 9th International Workshop on Joint Cloud Computing, JCC 2018
PB - Institute of Electrical and Electronics Engineers Inc.
Y2 - 26 March 2018 through 29 March 2018
ER -