reiserfs3 and -o sync revisited

reiserfs3 and -o sync revisited

[Matthias Andree]

Hi,

I have reiserfs v3.6 (as of 2.4.19) with the following patches:

A-00-version.diff.gz
A-01-relocation-4.diff.gz
A-02-commit_super-9-relocation.diff.gz
A-03-data-logging-25.diff.gz
A-03-logging-export.diff.gz
A-04-write_times.diff.gz
A-05-search_reada-4.diff.gz
B-02-gcc31_cleanup.diff.gz
B-03-help_cleanup.diff.gz
B-04-item_alignment_fix.diff.gz
B-05-reiserfs_make_bad_inode.diff.gz
B-06-bitops.diff.gz
C-bridge-netfilter.diff.gz
D-patchomatic.diff.gz
E-rmap14b.diff.gz
F-config.diff.gz

The A and B patches are Chris Mason's and the official Namesys patches,
possibly gzipped, the rest does not touch ReiserFS code except possibly
rmap.

I have run some bonnie++ 1.02c tests on an old swap partition I
converted to reiserfs, ext3fs or ext2fs, and I gather that using the
mount -o sync option makes virtually no difference on reiserfs, but
brings ext3 and ext2 nearly to a halt. Is mount -o sync implemented and
working?

Hardware: Duron 700, machine was NOT idle, VIA 82C686, Maxtor 4W060H4 as
single master (60 GB, 5400/min, 2 MB cache, write cache off), no slave
in same channel, hdc/hdd also idle.

ext2a: ext2   ext2d: ext2 dirsync   ext2s: ext2 sync
ext3_d=o: ext3 data=ordered ext3_d=w: ext3 data=writeback
ext3_d=j: ext3 data=journal ext3s_d=j: ext3 data=journal,sync
reiser: reiser no options, reiser_s: reiser sync
reiser_d=o: reiser data=ordered reiser_d=w: reiser data=writeback.

(I know the test set is not complete).

Note the data are from a single run of a workstation under my regular
use, but the drive (hda) I made the test on was idle otherwise.

Should I re-run this test with 2.4.20-rcN before anyone has a look at
this?


Version  1.02c      ------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
ext2a          360M            5483   6  3187   4           28930  23 112.5   1
ext2d          360M            4974   4  3310   4           23403  17  97.2   1
ext2s          360M             191   0   637   1           30690  21 113.5   1
ext3_d=o       360M            4898  13  2826   3           20767  15 114.0   1
ext3_d=w       360M            4518  14  3119   4           22996  19  92.6   1
ext3_d=j       360M            1930   7  1876   4           26598  22 101.8   1
ext3s_d=j      360M             222   1   219   0           29708  24  89.4   1
reiser         360M            5845   7  3173   3           28928  14 133.8   1
reiser_s       360M            5861   7  3103   3           28173  16 127.1   1
reiser_d=o     360M            5555   6  3253   3           30279  17 120.8   1
reiser_d=w     360M            5641   6  3223   2           31471  17 120.4   0
                    ------Sequential Create------ --------Random Create--------
                    -Create-- --Read--- -Delete-- -Create-- --Read--- -Delete--
files:max:min        /sec %CP  /sec %CP  /sec %CP  /sec %CP  /sec %CP  /sec %CP
ext2a 10:10000:0/23 12141  97 +++++ +++  1266   4 12307  97 +++++ +++  1214   7
ext2d 10:10000:0/23  2073  15 +++++ +++  1923   7  2185  17 19844  93  1605   8
ext2s 10:10000:0/23    17   0  4577  28    21   0    17   0   351   2    19   0
ext3_ 10:10000:0/23   951  26 19579  93  2045  19   792  18 19435  98  2689  33
ext3_ 10:10000:0/23   681  19 18572  92 10674  93   769  18 17919  94  7238  96
ext3_ 10:10000:0/23   330   9  9302  49 11259  97   337   9  1325   7  7620  95
ext3s 10:10000:0/23    15   0 18928  90    21   0    15   0 14455  77    16   0
reise 10:10000:0/23   247  47   234   1  5245  96   425  37    85   0   298   6
reise 10:10000:0/23   406  47   201   1  2899  52   413  35    91   0   318   7
reise 10:10000:0/23   431  62   219   1  2174  39   408  30    74   0   446  10
reise 10:10000:0/23   354  73   210   1  2351  41   732  72    76   0   350   7


-- 
Matthias Andree

Re: reiserfs3 and -o sync revisited

[Oleg Drokin]

Hello!

On Fri, Nov 15, 2002 at 12:07:09PM +0100, Matthias Andree wrote:
> I have reiserfs v3.6 (as of 2.4.19) with the following patches:
> A-01-relocation-4.diff.gz
> A-02-commit_super-9-relocation.diff.gz
> A-03-data-logging-25.diff.gz
> A-03-logging-export.diff.gz
> A-04-write_times.diff.gz
> A-05-search_reada-4.diff.gz

> The A and B patches are Chris Mason's and the official Namesys patches,
> possibly gzipped, the rest does not touch ReiserFS code except possibly
> rmap.
> I have run some bonnie++ 1.02c tests on an old swap partition I
> converted to reiserfs, ext3fs or ext2fs, and I gather that using the
> mount -o sync option makes virtually no difference on reiserfs, but
> brings ext3 and ext2 nearly to a halt. Is mount -o sync implemented and
> working?

-o sync is implemented by VFS.
I remember that some early implementation of data logging patches have broken
sync behaviour and reiserfs got huge boost in some tests ;) )
I do not know which version of the patches you aer using, though.

Chris, do your latest patches fix that sync problem?

Bye,
    Oleg

Re: reiserfs3 and -o sync revisited

[Chris Mason]

On Mon, 2002-11-18 at 03:10, Oleg Drokin wrote:
> Hello!
> 
> On Fri, Nov 15, 2002 at 12:07:09PM +0100, Matthias Andree wrote:
> > I have reiserfs v3.6 (as of 2.4.19) with the following patches:
> > A-01-relocation-4.diff.gz
> > A-02-commit_super-9-relocation.diff.gz
> > A-03-data-logging-25.diff.gz
> > A-03-logging-export.diff.gz
> > A-04-write_times.diff.gz
> > A-05-search_reada-4.diff.gz
> 
> > The A and B patches are Chris Mason's and the official Namesys patches,
> > possibly gzipped, the rest does not touch ReiserFS code except possibly
> > rmap.
> > I have run some bonnie++ 1.02c tests on an old swap partition I
> > converted to reiserfs, ext3fs or ext2fs, and I gather that using the
> > mount -o sync option makes virtually no difference on reiserfs, but
> > brings ext3 and ext2 nearly to a halt. Is mount -o sync implemented and
> > working?
> 
> -o sync is implemented by VFS.
> I remember that some early implementation of data logging patches have broken
> sync behaviour and reiserfs got huge boost in some tests ;) )
> I do not know which version of the patches you aer using, though.
> 
> Chris, do your latest patches fix that sync problem?

-o sync needs a little help from the filesystem, and my patches don't
add it.  Since we support fsync(file) after rename triggering a commit
on the rename, we don't really need -o sync.

But yes, I did fix my O_SYNC/fsync bug in the early versions of the data
logging patches ;-)

-chris

Re: reiserfs3 and -o sync revisited

[Matthias Andree]

Chris Mason <mason@suse.com> writes:

> -o sync needs a little help from the filesystem, and my patches don't
> add it.  Since we support fsync(file) after rename triggering a commit
> on the rename, we don't really need -o sync.
>
> But yes, I did fix my O_SYNC/fsync bug in the early versions of the data
> logging patches ;-)

So reiserfs ignores -o sync without complaining it's unimplemented. No
wonder reiserfs has an "eats-your-data" reputation.

NB: I find this inacceptable. Either have it reject chattr +S and mount
-o sync or fix it. But do not pretend it would be working when in fact
it's broken.

If this bug be unfixed by the end of the month, I'll have to convert any
r/w reiserfs systems left over to ext3fs.

The Linux file system semantics issue that has been going on for years
now is simply embarassing and awkward, be it the "dirasync" behaviour on
ext[23] or this ignoring -o sync. Linux needs to grow up. Seriously. And
quick.


I had also hoped that -o dirsync for ext2/ext3 would make it for 2.4.20,
but looking at Marcelo's BK tree, it hasn't. :-(

-- 
Matthias Andree