Home file server build, where do I start?

[Zarathustra[H]]

That sounds ridiculous - well, the "Buy a 5 day Drobo S..." part.

Leaving all DIY/prebuilt, etc. considerations aside, why not just buy a normal NAS in that case? (Qnap/Synology)?

Because normal NAS's suck in comparison. No flexibility. You have to have all the same drives.

Drobos are simply life changing.

Simply pop a drive out or in as you please whenever you please and it figures everything out on its own. Mix and match drives.

Last weekend my drive setup looked like this:

bay 1: 2tb wd green
bay 2: 2tb wd green
bay 3: empty
bay 4: 1tb wd black
bay 5: 2tb wd green

Yesterday I got my new 3TB WD green drive and simply popped into the empty slot. The Drobo automatically rebuilt the array and added the space without even having to restart. The space is available instantly. No re-configuring, no worrying about getting same size/brand/whatever drives. No issuing commands. Simply pop in the drive and you're done. Just stick whatever you have whenever you have it in the drobo and it just works.

If all my bays were full, I could have simply popped the smallest drive (my 1tb) out, and replaced it with the 3TB drive and it would still have made the space instantly available and then spent some time rebuilding the array, and managed itself.

Drobo's really are all that. They make traditional RAID5 arrays seem like the stone age.

After having a Drobo, I could never go back to traditional RAID5/FreeNAS/Anything else currently available. The Drobo is the storage Messiah and makes everything else seem like an obsolete pain in the ass by comparison

I was skeptical at first, but after seeing it in action, I was an instant convert.

It even comes with complete linux utils:

 sudo drobom info

---------------------------------------------------------
Drobo Name: Drobo S    Devices: /dev/sdb
      Time: Tue Jun 21 22:42:12 2011
---------------------------------------------------------
Configuration maxima: slots: 5, luns: 16, lunsize: 36028805 TB
Capacity (in GB):  used: 1287, free: 3660, total: 4947
protocol version:  0.11
SCSI emulation information:
Vendor: Drobo     Model: Drobo3           Revision: 3.00
/dev/sdb: bus=scsi4 channel=0 id=0 lun=0 (vendor=Drobo   , version=30527)
query slotinfo result:  number of slots: 5
slot   GB                Model               Status
   0 2000 WDC WD20EARS-00MVWB0                green
   1 2000 WDC WD20EARS-00MVWB0                green
   2 3000 WDC WD30EZRS-00J99B0                green
   3 1000 WDC WD1001FALS-00J7B1                green
   4 2000 WDC WD20EARS-00MVWB0                green
Firmware: 2.1.0
Revision: 5.28 ( 33144 ) built: Dec  7 2010,11:10:01
Features: NO_AUTO_REBOOT,NO_FAT32_FORMAT,USED_CAPACITY_FROM_HOST,DISKPACKSTATUS,ENCRYPT_NOHEADER,CMD_STATUS_QUERIABLE,VARIABLE_LUN_SIZE_1_16,PARTITION_LUN_GPT_MBR,FAT32_FORMAT_VOLNAME,SUPPORTS_NEW_LUNINFO2,feature x0800,LUN_MANAGEMENT,feature x2000 ,SUPPORTS_OPTIONS2,SUPPORTS_SHUTDOWN,leftovers (0x1f0000)
/dev/sdb /home Drobo S 26% full - ([], 0)
query options result:
        UseStaticIPAddress False
        UseManualVolumeManagement False
        YellowThreshold 85
        SpinDownDelayMinutes 15
        RedThreshold 95
        SpinDownDelay True
        NetMask 0.0.0.0
        DualDiskRedundancy True
        IPAddress 255.0.0.0
lun     size (GB) used    PTFmt  FStype:
  0     8796      1287      GPT ['EXT3']
---------------------------------------------------------

My Drobo is my Jesus array. Nothing else I have seen comes even close to comparing.

 

[john4200]

My Drobo is my Jesus array. Nothing else I have seen comes even close to comparing.

Do you have any benchmarks? Even a sequential read & write benchmark with dd would be interesting. I am wondering how it compares in speed to unRAID or FlexRAID, which provide the same functionality.

 

[Zarathustra[H]]

Do you have any benchmarks? Even a sequential read & write benchmark with dd would be interesting. I am wondering how it compares in speed to unRAID or FlexRAID, which provide the same functionality.

Good point. Speed has never been the Drobo's big selling point, and it wasn't a major concern for me when buying it, as I figured it would be bottlenecked by being on the network anyway, but that being said, that doesn't mean it can't perform.

Unfortunately - however - I have never seen any good - well structured - scientific style tests on this subject.

The best I have to reference is this guy's thread over on the droboforums, but it is of so-so value, as it tests different solutions with different drives in them, so they are not apples-to-apples.

I'd run you my own tests, but I am reluctant to do so right now, as mine is running in USB mode right now, which significantly bottlenecks it. I can confirm that it is fast enough to saturate the ~26MB/s max of the USB 2 controller though

This is why I plan on building a new server. My current server is one of those cheap Dell Zino HD rigs with an AMD780G chipset in it. I was unaware of this when I bought it, but apparently there is some sort of hardware bug in the eSATA implementation in the AMD 780 south bridge that prevents Port Multiplier from working properly under Linux, and the machine lacks any expansion slots, so I can't add a 3rd party controller to it.

Hindsight being 20-20 I should just have built a system like I usually do, but at $200 on sale the little cube just seemed like a fast and easy solution, which I am now regretting.

If you are really interested, I'm not opposed to disconnecting the Drobo from the server and hooking it up to my desktop or laptop for quick benchmark. It will likely have to be this weekend, late at night after everyone's gone to bed, as a lot of people in the house use the server and depend on it being up. I could even run HD Tach if that interests you.

 

[ChrisBenn]

Because normal NAS's suck in comparison. No flexibility. You have to have all the same drives.
Drobos are simply life changing.
Simply pop a drive out or in as you please whenever you please and it figures everything out on its own. Mix and match drives.

What advantage would buying a server + hooking up a drobo have over a server running flex-raid or unRaid (as john mentioned)?

Doesn't drobo offer a nas directly? Why would you build a server and then hook up direct attached storage to make a nas? I think this is what confuses me the most - you have do deal with the complexity/effort of a diy server (albiet with some of the disk management functions abstracted by the drobo) with the poor performance and expense of the drobo. It seems like the worst of both worlds.

If being able to mix and match random size drives is your #1 criteria then you have pretty much limited yourself to drobo or flex/unRaid.

But really, with the extra expense of the drobo you could have just bought a matched set of cheap 2TB drives and a qnap or synology nas (if it's simplicity you want) - and it would be a cheaper solution with greater performance.

I guess in the end 2TB hard drives are so cheap I don't see any point in trying to mix and match old 500gig, 1tb, etc. drives. Especially when the entry level cost to doing so outstrips the cost of buying new drives.

It seems to me the decision would be something like

1) Use mix and match of hard drive sizes -> (yes) drobo/flex/unraid (no) everything
2) DIY or rolled nas
3) Is bitrot protection desired for my application (basically do I need zfs)
4) what type of expansion options do I need (single disk, groups, online, etc.)
5) cost
6) performance

I just don't see the DAS Drobo + Server option ever being a valid choice out of that matrix. A drobo NAS maybe, but not a DAS drobo + server.

 

[Zarathustra[H]]

I just don't see the DAS Drobo + Server option ever being a valid choice out of that matrix. A drobo NAS maybe, but not a DAS drobo + server.

To each their own. I don't see the performance being an issue, besides I don't think its that bad. Wait until I have benched it with modern drives. The simplicity of the thing, and the flexibility for future upgrades is what really drew me to to it.

As far as why I got a Drobo S and attached my own server, vs. the Drobo FS (their NAS) its simple. a full fledged Linux server has much more flexibility and options than a integrated NAS running on low end ARM processors.

 

[john4200]

To each their own. I don't see the performance being an issue, besides I don't think its that bad. Wait until I have benched it with modern drives. The simplicity of the thing, and the flexibility for future upgrades is what really drew me to to it.

My understanding is that you are locked-in to Drobo since they use a proprietary disk format, with some sort of striping.

With UnRAID, FlexRAID, or SnapRAID/mhddfs (not exactly the same as the others since no real-time parity generation), each of your data drives is readable by itself with pretty much any linux distro. I think that is a great feature for flexibility, portability, and longevity.

The reason I was asking for a benchmark is because performance would seem to be the only possible reason why Drobo might be a better choice than UnRAID or FlexRAID. I'm not sure if Drobo does have a speed advantage, but since my understanding is that Drobo does something like RAID 5 striping, it conceivably MIGHT have a speed advantage over UnRAID or FlexRAID, which are more like RAID-4 (but on a file level), and so do not have the speed benefits of RAID-5-like striping.

 

[Zarathustra[H]]

The reason I was asking for a benchmark is because performance would seem to be the only possible reason why Drobo might be a better choice than UnRAID or FlexRAID. I'm not sure if Drobo does have a speed advantage, but since my understanding is that Drobo does something like RAID 5 striping, it conceivably MIGHT have a speed advantage over UnRAID or FlexRAID, which are more like RAID-4 (but on a file level), and so do not have the speed benefits of RAID-5-like striping.

I can certainly bench the Drobo S, but I have no experience UnRAID or FlexRAID, so I'll hvae to leave that for someone else.

I will get back to you with the bench.

--Matt

 

[john4200]

I can certainly bench the Drobo S, but I have no experience UnRAID or FlexRAID, so I'll hvae to leave that for someone else.

Thanks, I did not mean for you to bench anything but the Drobo.

FlexRAID and UnRAID are basically limited to the speed of a single drive (in fact, slower than a single drive on write, since there is some overhead with the real-time parity generation).

So the question is whether Drobo is faster than a single drive, and if so, how much?

 

[Zarathustra[H]]

Thanks, I did not mean for you to bench anything but the Drobo.

FlexRAID and UnRAID are basically limited to the speed of a single drive (in fact, slower than a single drive on write, since there is some overhead with the real-time parity generation).

So the question is whether Drobo is faster than a single drive, and if so, how much?

My guess is as follows, but I will follow up with data:

Faster than a single drive in sustained read up to a certain level where the ARM processor starts bottlenecking it.

Writes will likely be slightly slower than a bare drive for the same reason as you state above.

I'll bench it and let you know though.

 

[Zarathustra[H]]

Hmmm... I'm having some trouble with the eSATA connection on my new desktop now. (worked perfectly on my old X58 system). I need to poke around and try to fix this.

I'll post some benches when I can though.

 

[treadstone]

@john4200: FlexRAID isn't slower than a single drive, it is only as fast as the drive you are putting the data on as there is no parity generation at the time of the actual write to the disc (in snapshot RAID mode that is). So the throughput is only as fast as the disc you write to. Can't talk to any of the other systems as I don't use them.

 

[john4200]

@john4200: FlexRAID isn't slower than a single drive, it is only as fast as the drive you are putting the data on as there is no parity generation at the time of the actual write to the disc (in snapshot RAID mode that is).

I was talking about real-time version of FlexRAID.