The 3TB debug thread :-)

[Rolf Larsson]

Hi Joao,

I now have a DNS-323 (rev A1) flashed with standard 0.1RC2 and two Seagate 3TB disks in it, at your disposal for a week or two. I'm more than happy to try stuff out on them. I may even be able to arrange ssh and http access for you, if that would aid you. What do you want me to try out, specifically?

I'm fluent with Linux, but haven't had the need to fiddle much with disk partitioning the past five years or so.

[Joao Cardoso]

On Wednesday, October 3, 2012 10:17:02 PM UTC+1, Rolf Larsson wrote:

Hi Joao,

I now have a DNS-323 (rev A1) flashed with standard 0.1RC2 and two Seagate 3TB disks in it, at your disposal for a week or two.

Thanks a lot Rolf!

 

I'm more than happy to try stuff out on them.

To simplify, please start removing all Alt-F packages from the box. This will save us a lot of headaches.

Them download the attached files to /usr/www/cgi-bin and make them executable. They affect the Disk Wizard and the Disk Partitioner web pages and they will *not* survive a reboot (please remember this, or you will be really using and testing the RC2 versions of the files)

I would then start using the Disk Wizard to create all flavors of disk layouts, starting with the "One big filesystem per disk, for easy management (standard)", as it is the fastest. After partitioning finishes you can see if all the disk space is in use either in the Status or the Disk Filesystem pages. Also take a look at the Disk Partitioner to see if the free space is residual.

You could then try to use the Disk Partitioner to partition the disk with several sized partitions, verify that the partition is successful and using all disk space, and create filesystems on them using the Disk Filesystem web page.

You could then try to create several RAID flavours using the Disk Wizard. At least a RAID1, that is the most used layout.

Optional: You can then try to reproduce Saya report in the "DNS-323, current Alt-F firmware, and 3TB drives" thread, that states that the full 3TB disk capacity can't be used. To avoid lengthy file transfers to the box, you can just copy a big file under different names, such as

base=/mnt/sda2

dd if=/dev/zero of=$base/foo count=1000000

for i in $(seq 1 5400); do

cp $base/foo $(mktemp -p $base)

done

This test is going to take a veeery looong time to accomplish, and to be comprehensive it should be done on a RAID0 (faster) or JBOD, filling a huge ~6TB filesystem :-O

But I understand if you only do it on a "trivial" ~3TB filesystem ;-)

 

I may even be able to arrange ssh and http access for you, if that would aid you. What do you want me to try out, specifically?

I'm fluent with Linux, but haven't had the need to fiddle much with disk partitioning the past five years or so.

No problem, I just want to test the GUI. You know what you have to post  in case of problems, 'sgdisk -p /dev/sdx', 'sfdisk -luS /dev/sdx', 'df', 'dmesg', 'logread'...

Thanks a lot!

[Scott Vincent]

One more question, what model of Seagate 3TB drives did you buy and have been found to be reliable?

[Rolf Larsson]

My DNS-323 reports them as ST3000DM001-9YN166.

[Rolf Larsson]

As for reliable, they've been running for about a month. Not much to go on. However, I've been running Seagate Barracudas in them since 2007 without any problems. Much better than anything I've run before or since.

...I was waiting to post this until finished, but since I'm here anyway...

OK, I finally got some time to try this out. 

Test 1) Wizard - Both disks selected - One big filesystem per disk - ext4

Note: When starting, right disk (sda) was GPT partitioned, left (sdb) was MBR.

OK to start. 

Complains during procedure:

"Stopping all services and disks... done.

GPT partitioning disk sda... done.

GPT partitioning disk sdb... done.

fdisk: device has more than 2^32 sectors, can't use all of them fdisk: device has more than 2^32 sectors, can't use all of them

Creating and activating swap in disk sda... done.

Creating and activating swap in disk sdb... done.

Creating ext4 filesystem on sda2. It will take roughly 31 minute(s): xxx

While doing the above, the unit appears to have rebooted - I lost my SSH login, and I can't reconnect to the unit. The middle light (network indicator) blinks off and on ever 2-4 seconds. No disk activity whatsoever. Doesn't react at all to front button. On a reboot system is back to normal.

...then I remembered that the right-side LED is broken in this device, so it may still have been doing something. Ah well, I'll retry withone disk at a time in Test 2.

(I also noted that if you boot with a USB device in, it is mounted as /dev/sda - slightly confusing that the internal disks change letter. Logically, external devices should always start on sdc)

Test 2) Wizard - Right disk selected - One big filesystem per disk - ext4

Similar output (incl. warning about > 2^32 sectors). Works afterwards. Redid with left. Filling them up now.

[Joao Cardoso]

On Thursday, October 25, 2012 7:14:35 AM UTC+1, Rolf Larsson wrote:

As for reliable, they've been running for about a month. Not much to go on. However, I've been running Seagate Barracudas in them since 2007 without any problems. Much better than anything I've run before or since.

...I was waiting to post this until finished, but since I'm here anyway...

OK, I finally got some time to try this out. 

Test 1) Wizard - Both disks selected - One big filesystem per disk - ext4

Note: When starting, right disk (sda) was GPT partitioned, left (sdb) was MBR.

OK to start. 

Complains during procedure:

Not complains, progress reports :-)

 

"Stopping all services and disks... done.

"Stopping services" also means ssh , thus you will be kicked-off

GPT partitioning disk sda... done.

GPT partitioning disk sdb... done.

fdisk: device has more than 2^32 sectors, can't use all of them fdisk: device has more than 2^32 sectors, can't use all of them

This is a harmless fdisk diagnostic (to fix)

 

Creating and activating swap in disk sda... done.

Creating and activating swap in disk sdb... done.

Creating ext4 filesystem on sda2. It will take roughly 31 minute(s): xxx

While doing the above, the unit appears to have rebooted - I lost my SSH login, and I can't reconnect to the unit. The middle light (network indicator) blinks off and on ever 2-4 seconds. No disk activity whatsoever. Doesn't react at all to front button. On a reboot system is back to normal.

...then I remembered that the right-side LED is broken in this device, so it may still have been doing something. Ah well, I'll retry withone disk at a time in Test 2.

(I also noted that if you boot with a USB device in, it is mounted as /dev/sda - slightly confusing that the internal disks change letter. Logically, external devices should always start on sdc)

This is the linux kernel responsibility, not  Alt-F.

Device letters are assigned to disks by the kernel as soon as they appear; if a small USB disk is faster to spin-up then a large internal disk, and this often happens, then it will have a lower letter assigned to it.

That's why you can assign labels to filesystems (in Disk->Filesystems), to avoid this issue.

Test 2) Wizard - Right disk selected - One big filesystem per disk - ext4

Similar output (incl. warning about > 2^32 sectors).

But only the right disk is handled, right? i.e., only messages regarding it appears?

 

Works afterwards. Redid with left.

Good. Is the filesystem creation predicted time (31 minutes) correct? or near correct?

 

Filling them up now.

This means that you are not going to test the wizard with RAID1? :-(

Thanks for your testing! That's the only way for RC3 to see the light of the day.

[Rolf Larsson]

Sorry, I meant I'm filling the unit up with the test script you posted :-)

I've managed to fill the new unit to the brim (just doing one disk, I assume it'll work on the other one)

# df -h |grep sdb2

/dev/sdb2                 2.7T      2.7T         0 100% /mnt/sdb2

I also started to fill the remaining part of the JBOD file system on the one I did first. It's taking ages :-) So far, so good:

# df -h|grep md0

/dev/md0                  5.4T      4.5T    866.7G  84% /mnt/md0

Answers to your Questions:

> This is the linux kernel responsibility, not  Alt-F.

True, but I assume you can still 'lock' certain devices to specific designations which may be beneficial in this case, when the hardware is the same 'everywhere' - ie, if right bay is always SDA and left always SDB, there's less risk of confusing the two.

> But only the right disk is handled, right? i.e., only messages regarding it appears?

Yes, of course.

> Good. Is the filesystem creation predicted time (31 minutes) correct? or near correct?

Not entirely sure - in Test 1, it was still unresponsive after many hours, although it said it would take 31 minutes. In Test 2, I believe it may hve been something like 40 minutes for the first disk, no idea about the other one as I went to bed after kicking it off.

I'll keep an eye on it for the other ones.

> This means that you are not going to test the wizard with RAID1? :-(

Filling up = filling up with test script. See above :-)

I plan on doing RAID1 and JBOD, and perhaps RAID 5 (though I can only test that with an external smaller disk). However, this takes a nightmarish amount of waiting time. I'm almost wondering if I shouldn't have bought a more modern unit with better throughput =)

I'll keep you posted on the (slow) progress.

[Joao Cardoso]

On Saturday, October 27, 2012 7:49:13 PM UTC+1, Rolf Larsson wrote:

Sorry, I meant I'm filling the unit up with the test script you posted :-)

I've managed to fill the new unit to the brim (just doing one disk, I assume it'll work on the other one)

# df -h |grep sdb2

/dev/sdb2                 2.7T      2.7T         0 100% /mnt/sdb2

I also started to fill the remaining part of the JBOD file system on the one I did first. It's taking ages :-) So far, so good:

# df -h|grep md0

/dev/md0                  5.4T      4.5T    866.7G  84% /mnt/md0

 OK, you can stop doing that lengthy tests, everything seems to be OK,

Saya must have done something wrong with his filesystem.

Answers to your Questions:

> This is the linux kernel responsibility, not  Alt-F.

True, but I assume you can still 'lock' certain devices to specific designations which may be beneficial in this case, when the hardware is the same 'everywhere' - ie, if right bay is always SDA and left always SDB, 

No it is not. (reviewing the post: ah, you really mean uppercase sda?)

It happened to me once that sdb was the USB disk, sba the right and sdc

the left disk. We have no control over that, it is the kernel that

assigns it, and that depends on the spin-up time -- larger disks might

take several seconds to present themselves to the system

What I could do is to establish a link to the "right" device, say, I

could link /dev/sda to /dev/right, /dev/left to /dev/sdb, and then

always use /dev/right and /dev/left. This is done by some distributions

for CD/DVD, as its name is system dependent. But I can't link /dev/sdc

to /dev/usb, because you can connect several USB disks through a USB hub

to the box, and then what would I call  them?

The solution is to assign a label to each fillesystem, say sda2 will be Users, sdb2 Media, etc.

This issue has plagued the linux community for a while (it still does),

nowadays disks are used by its UUID or name, e.g.

root=/dev/disk/by-id/ata-ST3500413AS_6VMT70FK-part2

Network names suffer the same problem, e.g. eth0 and eth1 are not

guaranteed to always refers to the same physical network drive, so

'udev' has some rules for persistent devices names.

> there's less risk of confusing the two.

That's why a 'Bay' entry appears in all pages :-)

 

> But only the right disk is handled, right? i.e., only messages regarding it appears?

Yes, of course.

> Good. Is the filesystem creation predicted time (31 minutes) correct? or near correct?

Not entirely sure - in Test 1, it was still unresponsive after many hours, although it said it would take 31 minutes. In Test 2, I believe it may hve been something like 40 minutes for the first disk, no idea about the other one as I went to bed after kicking it off.

I'll keep an eye on it for the other ones.

 My "forecast" was done observing formatting times from my old 80GB

disks, so it is very likely that they are plain wrong for big

filesystems.

> This means that you are not going to test the wizard with RAID1? :-(

Filling up = filling up with test script. See above :-)

I plan on doing RAID1 and JBOD, and perhaps RAID 5 (though I can only test that with an external smaller disk). However, this takes a nightmarish amount of waiting time. I'm almost wondering if I shouldn't have bought a more modern unit with better throughput =)

I'll keep you posted on the (slow) progress.

 Thanks.

Remember, you can now skip the "filling the drive with data" test.

RAID is going to be slow, as the sync process typically runs at 15GB/s

(or is it 50GB/s?), and 5.4TB takes... hmmm, not enough fingers :-)

I attach (*) the fix to the "fdisk: device has more than 2^32 sectors, can't

use all of them" error message. Do you mind trying it?

Also, please

watch the Disk Partitioner, Filesystem Maintenance, RAID and Status page

for similar error messages.

(*) I'm having problems attaching files, I will wait until I receive the post by e-mail and I will then try to attach it.

[Joao Cardoso (Alt-F)]

On Monday 29 October 2012 15:13:24 Joao Cardoso wrote:

...

> I attach the fix to the "fdisk: device has more than 2^32 sectors, can't

> use all of them" error message. Do you mind trying it?

Trying to attach it by e-mail.

[Tom_1967]

How can I try it?