:Has anyone toyed with the idea of implementing a swap-based filesystem
:similar to Sun's tmpfs?
:
:Chuck Youse
I did it a couple of months ago. You simply use the VN device and
tell it to use swap as backing store, then newfs up a UFS filesystem
on it. You have the option to have it dynamically allocate and
deallocate swap, or you can force it to pre-reserve swap. See the
'vnconfig' man page and the -S option and the '-s reserve' option.
This is for -CURRENT only.
Generally speaking this isn't going to be as efficient as a real tmpfs
due to the update daemon syncing all filesystems every so often.
But if you pre-reserve the swap it *will* be just as efficient as a
normal filesystem. In fact, if you have multiple swap partitions
your tmpfs will wind up being interleaved and will have even better
performance. pre-reservation also gives you the ability to recover
the filesystem after a crash though for obvious reasons it can be
problematic to depend on this ability.
The big difference between MFS and VN is that MFS simulates a 'disk'
by using a special processes VM space, which means that disk blocks tend
to end up in memory twice - in the VM cache, and in the VM space. The
VN device implements the disk directly on a file or swap and does not
have the double-caching problem. MFS will tend to overload the VM
system if you use it to accomodate large files. VN operates more like a
normal disk, especially (in performance terms) when you use pre-reserved
swap as backing store.
vnconfig -e -s labels,reserve -S 100m /dev/vn0
disklabel -r -w vn0 auto
newfs -i 4096 /dev/rvn0c
tunefs -n enable /dev/rvn0c
mount /dev/vn0c /usr/obj
-Matt
Matthew Dillon
<[EMAIL PROTECTED]>
To Unsubscribe: send mail to [EMAIL PROTECTED]
with "unsubscribe freebsd-hackers" in the body of the message
