[GParted-Devel] MDRaid versus DMRaid on Recent GParted Live (debian SID based) Images

[Curtis G. <ge...@gm...>]

Hi Phillip,

I have taken the liberty of splitting this discussion into a more 
appropriately entitled email thread.

A brief description of the differences of the two GParted Live images 
(based on Debian Live - SID) are as follows:

0.19.0-beta1-2
--------------------
- GParted application compiled with --enable-libparted-dmraid and 
--enable-online-resize configure flags
- dmraid and kpartx packages from ubuntu (not debian)
- My Intel Software RAID recognized as Linux Software RAID device /dev/md126

0.19.0-beta1-3
- GParted application compiled with --enable-online resize
- dmraid and kpartx packages from debian
- My Intel Software RAID recognized as Device Mapper Raid device 
/dev/mapper/isw_efjbbijhh_Vol0

Responses to questions follow in-line:


On 14-06-09 11:35 AM, Phillip Susi wrote:
> On 6/9/2014 12:27 PM, Curtis Gedak wrote:
> | I agree that this finding is unexpected.  Following are some timings
> | using my Intel software RAID that contains two 160 GB SATA drives.
> |
> | Action A:  Shrink (not move) empty ext4 partition from 40 GiB to 20 
> GiB.
> | Action B:  Grow (not move) empty ext4 partition from 20 GiB to 40 GiB.
> |
> | GParted Live Image  Action   Time (mm:ss)   Partition Name
> | ------------------   ------   ------------ 
> -------------------------------
> |      0.19.0-beta1-2        A          0:06     /dev/md126p1
> |      0.19.0-beta1-3        A          0:01 
> /dev/mapper/isw_efjbbijhh_Vol01
> |      0.19.0-beta1-2        B          5:35     /dev/md126p1
> |      0.19.0-beta1-3        B          0:10 
> /dev/mapper/isw_efjbbijhh_Vol01
> |
> | These tests are on my primary computer that contains five drives and
> | almost as many operating system installations.  As can be seen from the
> | results, the operations on mdadm partitions are slower than dmraid
> | partitions.
>
> I wonder... can you look in the /sys/block entries corresponding to 
> the devices and see if their queue/optimal_io_size values differ?  
> Also can you show mdadm -D /dev/md126?

I looked through each the files in /sys/block/md126/queue on 
0.19.0-beta1-2 and compared the values to those in /sys/block/dm-0/queue 
on 0.19.0-beta1-3 and the values were identical.  If you need the exact 
files and values then let me know.

Output from mdadm command on 0.19.0-beta1-2 follows:

user@debian:/sys/block$ sudo mdadm -D /dev/md126
/dev/md126:
       Container : /dev/md/imsm0, member 0
      Raid Level : raid1
      Array Size : 156247040 (149.01 GiB 160.00 GB)
   Used Dev Size : 156247040 (149.01 GiB 160.00 GB)
    Raid Devices : 2
   Total Devices : 2

           State : clean
  Active Devices : 2
Working Devices : 2
  Failed Devices : 0
   Spare Devices : 0


            UUID : d24a487a:d7e2ede0:ace438e2:42a1f7c7
     Number   Major   Minor   RaidDevice State
        1       8       64        0      active sync   /dev/sde
        0       8       48        1      active sync   /dev/sdd
user@debian:/sys/block$

>
> | My testing of Debian has been within Virtual Machines.  The only 
> testing
> | I've done directly on bare metal is with Live images (CDs or USB flash
> | drives).  The above test with the GParted Live image should be
> | representative of Debian SID.
> |
> | Which Debian live image version would you like me to test?
>
> I don't think they maintain a live sid image, so I suppose testing of 
> the gparted iso will serve as a proxy.  You might try installing the 
> build of gparted from sid and check that.  Also are you sure that the 
> devices were named Vol01 and not Vol0p1 before running gparted? 

Yes, the partition is named Vol01 and not Volp1 when booting GParted 
Live 0.19.0-beta1-3.  Following is the "ls" output:

user@debian:~$ ls -l /dev/mapper
total 0
crw------- 1 root root  10, 236 Jun  9 21:11 control
brw-rw---- 1 root disk 254,   1 Jun  9 21:10 isw_efjbbijhh_Vol0
brw-rw---- 1 root disk 254,   0 Jun  9 21:10 isw_efjbbijhh_Vol01
user@debian:~$ ls -l /dev/dm*
brw-rw---- 1 root disk 254, 0 Jun  9 21:11 /dev/dm-0
brw-rw---- 1 root disk 254, 1 Jun  9 21:11 /dev/dm-1
user@debian:~$ ls -l /dev/md*
total 0
user@debian:~$

> I remember at one point we had the problem where the system made 0p1 
> and running gparted would make a duplicate 01 device.

When GParted is compiled *without* the "--enable-libparted-dmraid" 
configure flag, and an action is applied to the DMRaid device, GParted 
will still maintain both the 0p1 and 01 device entries. This behaviour 
has not changed.  If no actions are applied, then GParted does not 
create or delete any of these /dev/mapper entries.

>
> I'm working now on getting the dmraid and kpartx changes merged into 
> debian.

That's good news.  Hopefully we can get to the bottom of the performance 
differences so that this does not hinder Debian performance.

Curtis