1 |
On Wed, 7 Jan 2015 14:25:52 -0500, Rich Freeman wrote: |
2 |
|
3 |
> > Not as a general FS, but as a specific choice for $PORTAGE_TMPDIR it |
4 |
> > may be worth testing. XFS was designed for an environment that used |
5 |
> > temporary files that didn't need to be committed to disk, so its |
6 |
> > caching doesn't write to disk anywhere near as often. That means you |
7 |
> > would be working with files stored in RAM a lot of the time. |
8 |
> > |
9 |
> |
10 |
> If you're going to be saving the build files using mv/ln (or with cp |
11 |
> --reflink on a filesystem that supports this), then the LAST thing I |
12 |
> would do is use a different fs for PORTAGE_TMPDIR. The |
13 |
> best-performing option would be to make it a directory on the same |
14 |
> filesystem as wherever you're going to store the files by |
15 |
> moving/(ref)linking them. That makes the move/(ref)link operation |
16 |
> require minimal IO. |
17 |
|
18 |
That's not what I meant, but I see your point about using --reflink if |
19 |
making copies. My thought was to forget the whole tmpfs and copying |
20 |
think, set KEEP WORK in FEATURES and use XFS for PORTAGE_TMPDIR. That way |
21 |
most of the work is taking place in the cache without the frequent disk |
22 |
writes of other filesystems. I would expect XFS to be faster for this |
23 |
job, but have no data to support that assumption. |
24 |
|
25 |
Or you could just put TMPDIR on btrfs and snapshot after each emerge. |
26 |
|
27 |
|
28 |
-- |
29 |
Neil Bothwick |
30 |
|
31 |
Am I ignorant or apathetic? I don't know and don't care! |