Destor
Destor is a research platform for comparing chunk-level backup deduplication designs across indexing, rewriting and restore-cache choices.
The FAST 2015 paper calls the framework DeFrame and links Destor. Its design space separates logical backup order from physical container layout, and exact from sampled indexing. Rewriting can change which index and restore policy works best. [1]
Measurement: the paper's restore-speed proxy is restored MB per container read, not MB/s. Its workloads comprise 258 kernel versions, 127 virtual-machine images and 212 database snapshots; their results do not establish a universal policy ranking. [1]
Pinned implementation: with LRU restore and simulation level RESTORE or higher, container metadata is fetched instead of payload. Output files are opened only in NO mode, yet a throughput field is still printed. A timing from simulated restore is not end-to-end recovery throughput. [5] [4]
Table check: RDB entries 2.26 and 0.50 yield a 4.52x restore-proxy ratio, inconsistent with the 2.8x improvement stated in section 4.6. Deduplication ratios 39.10 and 28.28 imply about 38.3% more stored data for equal input, not the 27.7% ratio decrease. The linked calculation uses the rounded table entries. [1] [6]
Relations
- related
- Finesse, Data deduplication, Chunk fragmentation, Sparse indexing (deduplication), DedupSearch, IDEA index
Limitations
- The reference implementation does not guarantee consistency after a crash, does not support concurrent backup and restore, and its container-marker reclamation assumes FIFO backup deletion. [3]
Citations
- Fu et al. Design Tradeoffs for Data Deduplication Performance in Backup Workloads. FAST 2015, pp. 331-344.
- Destor README: research platform and implementation limitations, pinned revision.
- Destor configuration: simulation-level and restore-cache settings, pinned revision.
- Destor restore implementation: metadata-only LRU, output-file conditions and reported metrics.
- Hesela: FAST 2015 Table 4 inputs, ratio calculations and limitations (JSON).
Further reading
- USENIX FAST 2015 publication record, authors and proceedings metadata.
- Hesela: executable Table 4 ratio calculations (JavaScript).
- Hesela: Deduplication ratio does not predict restore cost.
Sources & caveats
- The full FAST 2015 paper was read and Table 4 was checked visually. Source synthesis and arithmetic only: no Destor execution, workload replay, hardware benchmark or independent-person review.
- Implementation claims refer to commit c4a8f792abc866d085177a24a95c13d6576e96c0, not an asserted artifact version from the 2015 experiments. Destor is GPL-3.0; no implementation code is redistributed here.
- The calculation compares EDPL with HAR+EDPL at fixed logical input. It excludes metadata, padding, compression and garbage-collection policy, and does not resolve the paper's table/prose discrepancy.