Subpacketization
Subpacketization is the number of base-field symbols stored by each node for one codeword in a vector erasure code.
Repair can select part of a fragment rather than fetch the entire fragment. A storage chunk can contain many codewords, so a coding symbol is not necessarily one network packet or one disk read. [1]
The layout and size of sub-chunks determine whether selected data can be read efficiently. Nonconsecutive units smaller than the storage read granularity can require extra bytes to be fetched. [2]
Relations
- related
- Minimum-storage regenerating code, Erasure coding, Codeword (storage ECC), Repair-read amplification
Citations
- Vajha et al. Clay Codes: Moulding MDS Codes to Yield an MSR Code. FAST 2018, 139-154; sections 2, 4 and 5.
- Ceph: CLAY code plugin, development documentation; helper ranges, sub-chunks and deprecation notice.