Skip to content

perf: set 4 MB readahead on BAM input (device-level or user-space) #254

@nh13

Description

@nh13

Observed that bumping block-device readahead from the Linux default (128 kB) to 4 MB sped up fgumi group on a large BAM by ~31% (33.96s → 23.38s, mean of 10 trials each, same binary, 6.8 GB input, page cache dropped).

Consider either (a) documenting read_ahead_kb = 4096 as a recommended setting for BAM-heavy runs, or (b) implementing user-space prefetch in the BAM read path so users get the benefit without needing root to tune the device.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type
    No fields configured for issues without a type.

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions