Deciding when to forget in the Elephant file system
Proceedings of the seventeenth ACM symposium on Operating systems principles
Space/time trade-offs in hash coding with allowable errors
Communications of the ACM
Metadata Efficiency in Versioning File Systems
FAST '03 Proceedings of the 2nd USENIX Conference on File and Storage Technologies
Parallax: managing storage for a million machines
HOTOS'05 Proceedings of the 10th conference on Hot Topics in Operating Systems - Volume 10
Hi-index | 0.03 |
While Copy-On-Write (COW) snapshot is the popular technique for online data protection, its first write request suffers from severe performance penalty because 3 I/Os are needed for each data block after creating snapshot. This paper proposes Pipelining-based COW (PCOW) method to minimize the snapshot impact on the first write request. When origin data is read from origin volume to a buffer queue, pending host write request can be serviced immediately. Origin data will be deferred to write to the snapshot volume in a background asynchronous thread. In order to demonstrate PCOW feasibility and efficiency, we have implemented both COW and PCOW in a standard iSCSI target as independent modules. We use popular benchmarks to quantitatively compare PCOW and COW techniques. Numerical results show that PCOW can effectively decrease first write penalty and improve performance.