Large Filesystem Woes
peterjeremy at optushome.com.au
Sat Jan 10 16:53:26 PST 2004
On Sun, Jan 11, 2004 at 12:25:46AM +0100, Dag-Erling Smørgrav wrote:
>Peter Jeremy <peterjeremy at optushome.com.au> writes:
>> Our main fileserver has a filesystem with 2.7e6 files and we
>> are continually running into undocumented "features" (aka bugs) as a
>> result of the large number of files.
>Is 2.7e6 a typo for 2.7e9? I can't imagine *any* modern file system
>having trouble storing barely three million files. My ~ alone has
>almost a million.
No, I am serious about the the 2.7e6. It's not so much a hard limit
as things like:
- A statement "these options are no longer necessary and will be
be removed in a future release" in the newfs(8)-equivalent man
page should read more like "these options are essential" (this
relates to dimensioning metadata allocation based on the expected
total number of files). The default values hit undocumented
metadata extent count limits at about 500,000 files. The table
showing suggested dimensioning guidelines only goes to 800,000 files.
- Failure to correctly dimension metadata during filesystem creation
will lead to "out of disk space" errors but there are no tools to
determine how close you are to this situation.
- Recommendations that a minimum of 5% free space should exist for
performing snapshots neglect to mention that the underlying code
apparently changes behaviour at about 12% free space.
- Metadata fragmentation can result in directories turning into
mode 0 files within a snapshot. (This is apparently more likely
to occur with less than 12% free space)
Admittedly, there has been a major upgrade of the FS code in a newer
OS release but my application requires the older release.
More information about the freebsd-hackers