RubikFS
RubikFS is a research read-only file system that groups similar data chunks before block compression to improve image compactness.
Layout: The FAST 2026 prototype extends EROFS with a userspace image builder and a modified kernel reader. It groups data by type, deduplicates chunks, separates hot data and clusters similar chunks before compression. This is a research extension, not a documented upstream EROFS feature. [1]
Recovery: Reordering chunks requires mappings back to their logical file positions. RubikFS retains file-to-chunk and packed-file location information; compressed data alone does not describe the original file layout. [1]
Evidence: The paper compares six images and three codecs. Its read-amplification study uses generated openEuler hot-data traces in FEMU/QEMU. Nominal block sizes are not equivalent: SquashFS uses pre-compression sizes, while RubikFS and EROFS use post-compression sizes. [1]
Trade-off: Sorting adds work and metadata, but total build time can fall if compression becomes faster. Reported gains depend on the image, codec, block definition and access pattern; they are not a universal startup-time or storage-saving guarantee. [1]
Read cost: In Table 2's 12% hot-data trace, LZMA reads 10.13 MB in 3.72 s versus LZ4's 16.41 MB in 1.21 s. Fewer bytes read do not necessarily mean a shorter run. [1]
Relations
Advantages
- Similarity and hotness grouping let image construction consider both compression and unnecessary reads. [1]
Limitations
- Hotness-grouping benefits depend on how well the supplied trace represents later accesses. [1]
Citations
Sources & caveats
- Primary-paper review, including rendered Tables 2-3; Hesela has not reproduced the prototype measurements or verified production deployment.
- Equation 1 prints min(4KB, Block Size / 16), whereas adjacent prose describes a 4 KB lower bound. No chunk-size formula is inferred here; this is not an author-confirmed erratum.