I have observed that traversing across file system trees and also "directly" accessing data by file name can greatly vary if the concerned file system is not fresh but aged. It seems, complex data structures like B*Trees or B-Trees get sub-optimal over time. I would suggest adding a test to the suite to reflect on these differences. In my personal tests with reiserfs (which I have considered stable and useful over decades) traverse time multiplied by factor 10. With btrfs and snapshots present, access may get laggy (the time till the first byte of a cluster gets read may extend to seconds). But even with less complex structures one might observe more or less small delays (e.g. direct access on ext4 by file name in deep file system trees).
Announcement
Collapse
No announcement yet.
Aging file systems
Collapse
X
Comment