[aklug] Re: File system thoughts

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Christopher Howard wrote:
> Just a thought I was pondering as I worked on my Operating Systems homework:
>
> We've all heard people say what their preferred file systems are. When
> we want help deciding for ourselves, the first thing we look for is
> those nice performance studies online, where multiple file systems are
> made to perform the same set of activities, and the speed results are
> lined up next to each other in order to determine strengths and weaknesses.
>
> But the only thing I never see discussed in these studies is the matter
> of adjustable file-system options -- allocation options in particular.
> For example, if you read through MKFS.XFS(8) you can see options for
> setting block size, inode size, allocation group count, et cetera et cetera.
>
> I'm no file systems expert, but it seems to me like it is one thing to
> judge a file system with default settings, but another matter altogether
> to see how it does after you've tuned it to your needs. One thing I'd
> like to do in future system installs is to learn more about the file
> system and to see how the can be manipulated in favor of the purpose of
> the partition.
>

Wow, this is a quiet stretch. With all the system admins on this list, I
figured just using the phrase "file systems" in an e-mail would get at
least two or three diatribes.

Well anyway, I been able to learn a few things, at least in relation to XFS:

1. You can trade off file capacity for performance (or vice versa) by
adjusting the block size. Larger block size can provide better
performance at the cost of storage. (Though, unfortunately, it would
seem that under Linux block size cannot be larger than the system page
size.)

2. You can improve file system throughput by using more allocation
groups. Under XFS, the file system is divided into isolated allocation
groups, which can each receive a distinct i/o stream simultaneously,
allowing for parallelization of the i/o within the same file system.

3. You can adjust inode size, though I'm not quite clear yet on what the
performance effects of that would be.

4. You can activate the "lazy-count" setting, which can improve
performance under intensive workloads, though this carries the risk of
"partial journal transitions" in the event of a power failure.

- --
 ________________________________
/ \
| Christopher Howard |
| linuxprogrammingforums.com |
| indicium.us |
| theologia.indicium.us |
\________________________________/
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v2.0.11 (GNU/Linux)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org

iEYEARECAAYFAkurJ6IACgkQQ5FLNdi0BcWjlQCgkxLKGSPrHU2cvIqaxgzEfdcx
bPAAnifAL1BlSjGsizVCbudfZDF3AbUT
=uJ8P
-----END PGP SIGNATURE-----
---------
To unsubscribe, send email to <aklug-request@aklug.org>
with 'unsubscribe' in the message body.
Received on Thu Mar 25 01:04:11 2010