[OpenAFS] File systems on Linux, again.

[Jerry Normandin]

AFS on EXT3?  No there are Metadata issues.  EXT3 was intended for this.
I inherited a mess here that I am fixing.  My predecessor built is using
Ext3 for the /vicepa filesystems.  It takes a hell of a long time to
create,
Delete, or rename files.  I tested with Bonnie++... here are my stats:

Initial file performance baselines taken from ENG02 with 4096kbyte
cache:

read performance on disk is 6.7 x faster than AFS
read performance on nfs is 2.0 x faser than AFS

write performance is actually impressive.  file creation and deletion
are very slow on afs.

file creation and deletion times is horrible.  I will be working on a
solution to solve this.  Once file creation and deletion times are
comparable to NFS, then AFS should perform well.

Bonnie++ Benchmarks

ENG02	AFS:
Version  1.03       ------Sequential Output------ --Sequential Input-
--Random-
                    -Per Chr- --Block-- -Rewrite- -Per Chr- --Block--
--Seeks--
Machine        Size K/sec %CP K/sec %CP K/sec %CP K/sec %CP K/sec %CP
/sec %CP
eng02.dafca.loca 8G  5296  47  6115  40  4074  34 19518  63 22494   7
125.6   4
                    ------Sequential Create------ --------Random
Create--------
                    -Create-- --Read--- -Delete-- -Create-- --Read---
-Delete--
              files  /sec %CP  /sec %CP  /sec %CP  /sec %CP  /sec %CP
/sec %CP
                 16    23   1  7777  79    23   0    23   1  4546  66
18   0
eng02.dafca.local,8G,5296,47,6115,40,4074,34,19518,63,22494,7,125.6,4,16
,23,1,7777,79,23,0,23,1,4546,66,18,0



ENG02 Local Disk:

Version  1.03       ------Sequential Output------ --Sequential Input-
--Random-
                    -Per Chr- --Block-- -Rewrite- -Per Chr- --Block--
--Seeks--
Machine        Size K/sec %CP K/sec %CP K/sec %CP K/sec %CP K/sec %CP
/sec %CP
eng02.dafca.loca 8G 35438  95 42686  23 16027   6 13516  31 24135   4
604.7   1
                    ------Sequential Create------ --------Random
Create--------
                    -Create-- --Read--- -Delete-- -Create-- --Read---
-Delete--
              files  /sec %CP  /sec %CP  /sec %CP  /sec %CP  /sec %CP
/sec %CP
                 16  2046  77 +++++ +++ +++++ +++  2230  84 +++++ +++
1461  17
eng02.dafca.local,8G,35438,95,42686,23,16027,6,13516,31,24135,4,604.7,1,
16,2046,77,+++++,+++,+++++,+++,2230,84,+++++,+++,1461,17



ENG02  NFS:
Version  1.03       ------Sequential Output------ --Sequential Input-
--Random-
                    -Per Chr- --Block-- -Rewrite- -Per Chr- --Block--
--Seeks--
Machine        Size K/sec %CP K/sec %CP K/sec %CP K/sec %CP K/sec %CP
/sec %CP
eng02.dafca.loca 8G 11051  27 11126   3  8944   5 11179  27 11186   2
118.1   0
                    ------Sequential Create------ --------Random
Create--------
                    -Create-- --Read--- -Delete-- -Create-- --Read---
-Delete--
              files  /sec %CP  /sec %CP  /sec %CP  /sec %CP  /sec %CP
/sec %CP
                 16   934   4  6092  12  2951   6   949   4  7922  12
2359   5
eng02.dafca.local,8G,11051,27,11126,3,8944,5,11179,27,11186,2,118.1,0,16
,934,4,6092,12,2951,6,949,4,7922,12,2359,5

So I have to backup the vicepa filesystems.  Create a xfs filesystem.
And restore.  I've got 3TB of data to deal with.

Trust me you do not want to use ext3 for the /vicepa file system!

-----Original Message-----
From: openafs-info-admin@openafs.org
[mailto:openafs-info-admin@openafs.org] On Behalf Of Russ Allbery
Sent: Thursday, November 29, 2007 2:36 PM
To: openafs-info@openafs.org
Subject: Re: [OpenAFS] File systems on Linux, again.

"Smith, Matt" <matt.smith@uconn.edu> writes:

> After the recent thread "openafs upgrade from 1.4.1 to 1.5.7", and a
> review of a thread[1] from July, I'm wondering if there is a
definitive
> recommendation for which file system to use on Linux AFS file servers.
> Ext3, XFS, JFS, something else?

It shouldn't make much of a difference, so I think you're safe choosing
your file system on whatever basis you'd choose a file system for any
other file server.  We use ext3 because of the stability, reliability,
and
"center of the mainstream" support in the kernel, which we always
considered more important than a bit of additional speed, but your
mileage
may vary.

XFS is probably the next most common choice.

I would be very leery of ReiserFS.  It has nice features, but the
recovery
tools are fairly horrific.

--=20
Russ Allbery (rra@stanford.edu)
<http://www.eyrie.org/~eagle/>
_______________________________________________
OpenAFS-info mailing list
OpenAFS-info@openafs.org
https://lists.openafs.org/mailman/listinfo/openafs-info