dc.contributor.advisor |
Nam, Beomseok |
- |
dc.contributor.author |
Jin, Kibeom |
- |
dc.date.accessioned |
2024-01-25T14:13:29Z |
- |
dc.date.available |
2024-01-25T14:13:29Z |
- |
dc.date.issued |
2017-08 |
- |
dc.description.abstract |
Emerging non-volatile memory has opened new opportunities to re-design the entire system software stack and it is expected to break the boundaries between memory and storage devices to enable storage-less systems. Traditionally, B-tree has been used to organize data blocks in storage systems. However, B-tree is optimized for disk-based systems that read and write large blocks of data. When byte-addressable non-volatile memory replaces the block device storage systems, the byte-addressability of NVRAM makes it challenge to enforce the failure-atomicity of B-tree nodes. In this work, we present NVB-tree that addresses this challenge, reducing cache line flush overhead and avoiding expensive logging methods. NVB-tree is a hybrid tree that combines the binary search tree and the B+-tree, i.e., keys in each NVB-tree node are stored as a binary search tree so that it can benefit from the byte-addressability of binary search trees. We also present a logging-less split/merge scheme that guarantees failure-atomicity with 8-byte memory writes. Our performance study shows that NVB-tree outperforms the state-of-the-art persistent index - wB+-tree by a large margin. |
- |
dc.description.degree |
Master |
- |
dc.description |
Departmentof Computer Engineering |
- |
dc.identifier.uri |
https://scholarworks.unist.ac.kr/handle/201301/72228 |
- |
dc.identifier.uri |
http://unist.dcollection.net/jsp/common/DcLoOrgPer.jsp?sItemId=000002380559 |
- |
dc.language |
eng |
- |
dc.publisher |
Ulsan National Institute of Science and Technology (UNIST) |
- |
dc.rights.embargoReleaseDate |
9999-12-31 |
- |
dc.rights.embargoReleaseTerms |
9999-12-31 |
- |
dc.title |
NVB-tree: Failure-Atomic B+-tree for Persistent Memory |
- |
dc.type |
Thesis |
- |