Badblocks run time (Dreamplug + USB)

So I put my hands on two seagate external hard drives (2T), and I want to use them as a raid device. Prior using the drives, I like to run some checks on them, to see if they deserve my data to be put on them. I ran badblocks on both drives at the same time. I don't think, that it has any impact on the speed. To record how much time did it take to run badblocks on 2T drives connected through USB to my Dreamplug, I captured some outputs:

root@plugged:~# badblocks -b 1048576 -w -o badblocks.W1E2JW11 -s -v /dev/sdc
Checking for bad blocks in read-write mode
From block 0 to 1907728
Testing with pattern 0xaa: done                                
Reading and comparing: done                                
Testing with pattern 0x55:  77.39% done, 92:31:15 elapsed

root@plugged:~# badblocks -b 1048576 -w -o badblocks.W1E2GMMM -s -v /dev/sdd
Checking for bad blocks in read-write mode
From block 0 to 1907728
Testing with pattern 0xaa: done                                
Reading and comparing: done                                
Testing with pattern 0x55: done                                
Reading and comparing: done                                
Testing with pattern 0xff: done                                
Reading and comparing:   2.04% done, 172:41:55 elapsed

Badblock with these parameters tests 4 patterns: 0xaa, 0x55, 0xff, 0x00. The latter output shows the time required to get to the read-back phase of the second pattern. So let's say the read and write takes up around the same time, meaning we have the time for 5 iterations:

5x = 172
x = 34.4

And for doing all the 8 phases, you will need 275.2 hours, around 11.5 days, let's say 12 days. I am not sure, that this is the most economic way of doing it, because the drives are consuming a lot power, so it might make sense to drive them as fast as possible with a more powerful machine.

NFS installation

So I decided to put an NFS server to my DreamPlug. First, an easy step:

apt-get --no-install-recommends install nfs-kernel-server

Let's see what do I have:

root@plugged:~# lsmod | grep nfs

nfsd                  243695  11 
exportfs                3108  1 nfsd

So, next step is to export a directory. I would like to create a read-only export.

cat /etc/exports

/mnt/md2        192.168.0.2(ro,no_subtree_check)

TODO: Some security Okay, so let's restart the nfs server:

/etc/init.d/nfs-kernel-server stop
/etc/init.d/nfs-kernel-server start

Install the client on the other machine:

sudo apt-get --no-install-recommends install nfs-common

And mount the export:

sudo mount 192.168.0.4:/mnt/md2 /mnt/md2

It does not seem to perform very well. For example, from the client, through wifi:

time ls -la /mnt/md2/1/ > /dev/null

Gives

real 1m18.605s
user 0m0.560s
sys 0m2.632s

Whereas running the same stuff on the DreamPlug:

real 0m2.572s
user 0m1.650s
sys 0m0.920s

And that's fast. Let's try to dd some files.

dd if=/mnt/md2/1/ffc4f5bc7f6b78ac371f3ecebdd9701e4cb8c68f of=/dev/null

Gives:

577409024 bytes (577 MB) copied, 103.255 s, 5.6 MB/s

And what is the raw speed? On the server:

nc -l -p 4545 -q 1 < /mnt/md2/1/ffc4f5bc7f6b78ac371f3ecebdd9701e4cb8c68f

On the client:

nc 192.168.0.4 4545 | dd of=/dev/null

And the results:

577409024 bytes (577 MB) copied, 103.84 s, 5.6 MB/s

So nfs gives almost no overhead. (Please note, that the 1 sec was the waiting time added by the netcat server. That's negligable.) Let's see the same thing locally on the server:

dd if=/mnt/md2/1/ffc4f5bc7f6b78ac371f3ecebdd9701e4cb8c68f of=/dev/null

Result:

577409024 bytes (577 MB) copied, 19.4105 s, 29.7 MB/s

NFS over a gigabit connection (after umount-mount, to clear the cache):

577409024 bytes (577 MB) copied, 24.6491 s, 23.4 MB/s

Without NFS (nc way):

577409024 bytes (577 MB) copied, 20.9865 s, 27.5 MB/s

So yes, NFS has some overhead at these rates.

Dreamplug - Updated Debian - gpio messages

I updated my dreamplug to the latest debian, and I see these messages in dmesg:

the gpio timer is called ! time is 211

This is somewhat really annoying, so I decided to try to install a fresh debian. The instructions I used for the update are: Update guide

Dreamplug serial console, plus faster boot

Some months ago, I put my hands on a Dreamplug device. I haven't done too much with it, just put arch on it. Now I decided to log in again, and play around with it. First I need to log in with the serial console. In order to do that, I need to be able to use my serial (ttyUSB0) as a simple user. I first added myself to the dialout group:

sudo gpasswd dialout -a matelakat

And as I don't want to log on/off, I activated the group memberships:

newgrp dialout

And off you go, I just had to start minicom:

minicom

Let's make booting a bit faster. See here. I will make a note of my original bootcommand here (The whole stuff needs to be one line!)

bootcmd=setenv ethact egiga0; ${x_bootcmd_ethernet}; setenv ethact egiga1;
 ${x_bootcmd_ethernet}; ${x_bootcmd_usb}; ${x_bootcmd_kernel};
 setenv bootargs ${x_bootargs} ${x_bootargs_root};
 bootm 0x6400000;

And so it is modified to:

${x_bootcmd_usb}; ${x_bootcmd_kernel};
 setenv bootargs ${x_bootargs} ${x_bootargs_root};
 bootm 0x6400000;

With this:

Marvell>> setenv bootcmd '${x_bo...

And save the stuff:

Marvell>> saveenv

And reset

Marvell>> reset

So that the device will boot significantly faster.

Disable mdadm autodetect

I wanted to create a raid array from USB drives. As USB drives are changing, I don't want mdadm to auto-assemble the arrays during system startup.

[root@alarm ~]# cat /etc/mdadm.conf | grep DEVICE
DEVICE /dev/null

Dreamplug USB drive performance

I attached an USB drive to my new toy (dreamplug) to see what performance can I expect. Let me share the results:

dreamplug-debian:/mnt/disk# dd if=/dev/zero of=j bs=1024 count=1000000         
1000000+0 records in                                                           
1000000+0 records out                                                          
1024000000 bytes (1.0 GB) copied, 61.06 s, 16.8 MB/s                           
dreamplug-debian:/mnt/disk# dd if=j of=/dev/null bs=1024 count=1000000         
1000000+0 records in                                                           
1000000+0 records out                                                          
1024000000 bytes (1.0 GB) copied, 37.4651 s, 27.3 MB/s                         

I think I am happy with these results. The filesystem was reiserfs.