Drobo

Linux, 4TB drobo

I purchased a drobo after learning that it supported ext3. I bought four 1tb drives and connected the drobo usb to my linux host. The device reports two 2tb drives to my OS, and I get two independent devices in /dev. Linux thinks this is two independent 2TB drives. My questions are:

  1. I’d rather not have 2 independent devices, is it acceptable to increase the lun size for this device to 4tib to support the drives as a single device?

  2. According the the capacity calculator, inserting four 1tb disks should yield 2.7tb in usable storage, but the devices reports to linux it has 4tb available. If 4tb is truly available, where is the redundancy? If 2.7tb is all that is available, but linux sees the device as 4, what happens when data past 2.7tb is written to the drobo?

I would like to see a bit more detailed documentation.

Thanks in advance for the assistance,
Lane Maxwell

#1 maybe… mixed reports.
#2 is because of thin provisioning. Without thin provisioning you’d have to reformat the volume and migrate the data every time you expanded your storage.

This article explains both:
http://support.datarobotics.com/app/answers/detail/a_id/23

And this thread indirectly explains what happens as you approach the real capacity of the Drobo - it significantly slows the transfer rate to let you know you’re getting close to full.
http://www.drobospace.com/forums/showthread.php?tid=293

As far as anyone can tell, LUNS > 2 TiB do not work properly. DRI does not appear to be aware of the problem, although it has been reported to them several times. I think different issues get conflated, and they fix one issue, but other ones remain outstanding. There was a usb issue, and a firewire issue, which were both addressed by DRI getting fixes into the Linux kernel. All user tests indicate there is still another issue with large LUNS.

Here is a thread about various tests of various file systems, and the universal failure to get hoped for results:

http://groups.google.com/group/drobo-talk/browse_thread/thread/34f08da1708d2f7b?hl=en

from the README distributed with drobo-utils:

Multiple LUNS
LUN is an abbreviation of ‘Logical UNit’. The origin of the term is SCSI terminology. When RAID units became too large for support in the past, and were sub-divided to present smaller units the operating system. The default LUNSIZE on Drobos is 2 TiB (adjustable using the tools.) If more disk space (after allowing for parity/redundancy) than LUNSIZE is installed in a unit, Drobo will show a second (or even third) LUN. Each LUN shows up in Linux as a separate disk (examples if the first LUN shows up as /dev/sde, the next will be /dev/sdf, then /dev/sdg.)

It always reports an integral number of LUNS, each of the LUNSIZE. The space goes where you put it. For example, you leave LUN0 empty, and write to LUN1, then 2 TiB will go there, the entire file system can be filled without causing any problem.

Assuming the total amount of disk allocated to both luns gets close to the physical space available, you get do a condition where space used is > RedThreshold% full. Drobo will then slow down, and alert you.
It will be asking you to upgrade or add disks, so that you don’t over load the disks.

DRI has a decent knowledgebase. This forum has some nuggets. The google group (referred to above.) has a lot of linux-oriented threads, and some pages that are kind of a FAQ/General culture.

Thank you both for your replies, they were very helpful.

Since linux has large block device support available, it would be swell to see the drobo recognized as such and only report the actual space that is available. Perhaps this is where a drobo device driver would come in handy. I like the drobo, it’s a nifty device, but I don’t like the idea of of the device intentionally throttling write speeds to notifiy of nearing capacity.

Thanks again for the replies, I’ll continue to search for the optimum configuration for this cute little gadget.

Lane

one answer:

drobom set RedThreshold 99
drobom set YellowThreshold 95

Will adjust the alerting & slowdown thresholds (just out in this week’s release 0.6.1 of drobo-utils.)

another answer:

A wishlist item is to build a FUSE file system to just intercept the statfs call (the only one affected) to report actual physical usage. It should not be a big project… volunteer?