Document the softupdate stuff and also warn people against using

it on root unless root is very large.
This commit is contained in:
Jordan K. Hubbard 2001-03-11 04:24:52 +00:00
parent 46c9472cd6
commit abbc9c16f2
2 changed files with 34 additions and 0 deletions

View file

@ -109,6 +109,23 @@ slices. For FreeBSD partitions, you can also toggle the "newfs" state
so that the partitions are either (re)created from scratch or simply
checked and mounted (the contents are preserved).
If you set (S)oftUpdates on a filesystem, it will cause the
"Soft Updates" policy to be in effect for it. This basically causes
both metadata and data blocks to be written asynchronously to disk,
but with extra state information which causes the metadata and any
related data blocks to be committed in a single transaction. This
results in async metadata update speeds (which are considerably
faster than the default sync) without the potential for data loss
which could occur if you simply mounted the filesystem with purely
"async" update policy and then had a power failure. If you wish
to later turn the softupdates policy back off, use the command
"tunefs -n disable devicename". NOTE: It is probably not wise
to use this on your root filesystem unless you have a large
(e.g. non-standard size) root. The reason is that smaller filesystems
with significant activity can temporarily overflow if the soft updates
policy results in free'd blocks not being "garbage collected" as fast
as they're being requested.
When you're done, type `Q' to exit.
No actual changes will be made to the disk until you (C)ommit from the

View file

@ -109,6 +109,23 @@ slices. For FreeBSD partitions, you can also toggle the "newfs" state
so that the partitions are either (re)created from scratch or simply
checked and mounted (the contents are preserved).
If you set (S)oftUpdates on a filesystem, it will cause the
"Soft Updates" policy to be in effect for it. This basically causes
both metadata and data blocks to be written asynchronously to disk,
but with extra state information which causes the metadata and any
related data blocks to be committed in a single transaction. This
results in async metadata update speeds (which are considerably
faster than the default sync) without the potential for data loss
which could occur if you simply mounted the filesystem with purely
"async" update policy and then had a power failure. If you wish
to later turn the softupdates policy back off, use the command
"tunefs -n disable devicename". NOTE: It is probably not wise
to use this on your root filesystem unless you have a large
(e.g. non-standard size) root. The reason is that smaller filesystems
with significant activity can temporarily overflow if the soft updates
policy results in free'd blocks not being "garbage collected" as fast
as they're being requested.
When you're done, type `Q' to exit.
No actual changes will be made to the disk until you (C)ommit from the