Chopper
Chopper is a filesystem diagnostic framework that samples workload and environment factors to expose unexpected block-allocation layouts.
The FAST 2015 study used Latin hypercube sampling and sensitivity analysis with 16,384 experimental treatments on Linux 3.12.5. Its d-span response measures spatial spread in bytes, not elapsed time; a multi-file treatment contributes its largest file d-span. The tail across selected treatments is not a production request-latency percentile. [1]
Equal d-span can hide different internal layouts. The pinned monitor computes an outer endpoint difference for selected leaf extents. [2] In the linked synthetic example, two four-block files each span 32,768 bytes but have two versus four contiguous runs. No device latency is calculated. [4]
The author artifact includes workload generation, layout collection and analysis. Creating and mounting experimental filesystems requires a disposable environment; its historical setup is not a safe command to run against existing data. [3]
Relations
Limitations
- Results from the historical kernel do not establish defects in current ext4 or XFS.
- An allocation-layout proxy cannot by itself predict latency on a particular storage stack.
Citations
- He et al. Reducing File System Tail Latencies with Chopper. FAST 2015, pp. 119-133, sections 2.1-2.4 and 3.1.1.
- Chopper ef41fc432e4d: Monitor.py, get_d_span_from_extent_list and per-file d-span output.
- Chopper ef41fc432e4d: author tutorial, experimental setup and worker isolation.
- Hesela: exact synthetic layouts with equal spatial span and unequal contiguous-run counts.
Sources & caveats
- Source review and original synthetic arithmetic only; no Chopper benchmark, kernel patch or physical-device experiment was run.
- Example endpoints are half-open byte intervals. It illustrates information loss in a span metric, not a byte-for-byte replay of every historical filesystem parser.