Dear StorageMojo: cheap home bulk storage?

by Robin Harris on Monday, 8 August, 2011

Several readers have written in lately with roughly the same question: what’s the best way to build cheap home bulk storage?

Here’s how 1 writer put it:

I was hoping you could provide me with some advice. I have so many external drives that I have to swap. I have almost 10 Terabytes of data – mostly movies that I would like to consolidate into one “volume”. I was thinking about building a 5-bay JBOD Raid (a cheap enclosure) and was also thinking of using MacZFS to handle the storage pool part.

Price is important. Performance isn’t.

How would readers propose to do this on their favorite OS? The folks asking this question aren’t full time sysadmins – they already have their ideas – so let’s not get too esoteric.

The StorageMojo take
This question has been vexing me as well. I have a small Thunderbolt array – a Promise Pegasus R4 – hooked up to a new iMac for video editing. I’d like to reconfigure it from RAID 5 to multiple RAID 0 stripes for speed and capacity. The question is how to back up those vulnerable RAID 0 partitions.

It’s looking like large FireWire drives are the right answer. But maybe you have a better one.

Courteous comments welcome, of course.

{ 50 comments… read them below or add one }

anonymous August 9, 2011 at 1:07 pm

FreeBSD on a commodity motherboard with a bunch of multi-TB drives with a zfs raidz file system sitting on top.


Whatever you want for a server with a Drobo attached.

Duncan August 9, 2011 at 1:18 pm

Refurbs, all the way!

A few years back I bought a refurb Mac Mini (with server OS) from the Apple online store. I also bought two Qx2 4-disk RAID enclosures from Other World Computing (new in this case, but later bought a third unit refurbed) on a Black Friday deep discount, one filled with 2 TB Hitachis for speed and the other with 2 TB WDs (5400 rpm) for more economical backup. This was the earlier model Mini so I pried the top off and attached the primary (Hitachi) RAID-5 via direct SATA, good for 150MBs transfer, to one of the internal SATA connectors (I located a proper-gender SATA to eSATA cable online). The slower RAID is connected via FW-800, and backs up the primary using Time Machine. Boot the Mini with a monitor attached to trigger the proper video resolution, but after that run it headless.

(A newer Mini can be used with some judicious handiwork to sneak out a SATA connection, or simply use FW-800 for serviceable speed and off-the-shelf cabling.)

The benefit of this over a dedicated NAS unit:

– Native OS X file sharing (plus SMB and NFS)
– Vendor-supplied remote management
– Full suite of server applications (calendar, mail, address book, etc.) available if that’s useful
– Configuration-free RAID-5 (or other RAID levels if desired) using hardware switch on Qx2
– Network Time Machine backups for other Macs

Obviously this can be replicated, probably cheaper, using PC/Windows/Linux components. But for a Mac-only network this setup has proven to be very fast and reliable once the initial configuration is done. I think the most salient advice no matter which way one goes is to look at refurbished hardware. It carries the same exact warranty and I haven’t been able to find any cosmetic differences. You don’t get the fancy original packaging, however – just brown printed boxes – so there is that.

Anonymous August 9, 2011 at 1:22 pm

For Windows you should use windows home server with the built in software raid you are only limited to the amount of sata ports on the board.

For Mac / Linux I would do a similar setup that Backblaze has done. Something like FreeNAS.

Nathan August 9, 2011 at 1:26 pm

When I asked this question of myself six years ago, it seemed like the answer was to buy a RAID controller and set up a RAID 5. These days, it looks more and more like software RAID is much closer to the answer.

ZFS on OpenSolaris is certainly one way to go. Unfortunately, at least the last time I checked it wasn’t possible to add a drive to a zpool, so if you’re planning on growing your motley collection of drives over time then you will wind up with some awkward configurations. In any case, steer clear of the BSD and FUSE ports of ZFS, as they are reputedly less stable and much less supported.

Another solution, of course, is just to buy a Drobo and call it a day.

The middle path seems to be a standard software RAID 5 on your OS of choice (they should all support it). Backups can be done to the cloud or to large external drives (I can back up my 1.2TB RAID 5 to a single 2TB USB drive which I store offsite; the RAID lets me survive single drive failure and the weekly backup means even a catastrophe (or an accidental deletion) can’t wipe out too much). Keeping the RAID on a separate box and backing up my desktop and laptop through a cron job means that I have at least two and usually three copies of most stuff, and then large data (such as uncompressed photos) can be stored directly on the RAID for a redundancy hit.

It seems like for home storage you can have it cheap, easy, and reliable–choose two. If performance is a concern (i.e. if you’re using your RAID as primary storage) I suspect you only get to choose one.

RPRoss August 9, 2011 at 1:43 pm

Still a ZFS fan here. EON, Nexenta or Sol11E. The napp-it add-on can be very useful for non-admins. Something like 4x Hitachi 3TB in an HP MicroServer and the LSI 9212-4i4e for expansion. LSI internal is connected to 4x 2.5″ mobile rack for boot and cache drives if you want them. ~$1,500 from NewEgg. Pretty slick. Robert

David Magda August 9, 2011 at 3:23 pm

Perhaps we can get a a few budget ranges suggested? Or would storage range ranges be better? For instances one can get a Backblaze v2 pod for around US$ 2K and simply add drives as space is needed (up to 135 TB).

Assuming that IOps are not all that important, would the categories of: under $300, $301-700, $701-1500, over $1500, be appropriate? Or perhaps: under 3 TB, 3-7 TB, 7-12 TB, over 12 TB?

Otherwise it’s easy enough to throw out any old random configuration: either a DIY x86 machines, or pre-canned NAS devices (Netgear, Drobo, etc.).

The question “what should I get?” is best answered by “it depends”. The one writer you quote says “Price is important”; awesome, gives us a bloody number.

Logan August 9, 2011 at 7:05 pm

Hello everyone,
The original post was my question I asked Robin.

I was thinking of the TowerRAID 5-bay drive enclosure (USB version) with 5 3TB “green”/consumer hard drives.

So in all under $1000 (or around $1000) for 15 TB of usable space (cloud for backup).

Then using MacZFS or Ten’s Complement Z-410 to handle the “RAID”/storage pool part.

If anyone has other ideas in that price range (and usable storage capacity) please offer your advice. I’ve been reading all comments.

The Backblaze and building my one NAS with FreeNAS seems way to complicated no? And Drobo is too expensive on a price per gigabite comparison (and you loose the largest drive – for backup).

I’m not going to use a RAID-5 configuration and loose 1 drive, because I will have use a cloud back-up.

Remember, price is key, performance is not. I will not regularly write to the storage pool. It only store my data (mostly read-only) from basically all my family’s life (home vids, pictures, old software cd rips, etc…).

I chose ZFS because reliability is very important… will have all my family’s important stuff.

Logan August 9, 2011 at 7:09 pm

Also, upgradability/expandability is not important. When I need a new one (probably in 5+ years) i will make a new storage pool.

brian August 9, 2011 at 7:33 pm

a small mini-atx PC (the size of a breadbox) running either openfiler ( or freenas ( I don’t want to use an old tower, they are loud, can’t fit under a desk or dresser out of the way, and draw too much power..

Anthony August 9, 2011 at 8:00 pm

I use Solaris 11 Express
I have the old Asus M2N-PV board 65W Dualcore CPU 4GB RAM
Boot Drives 2x Seagate ES 250GB
Mirrored 1.5TB Drives with auto replace and auto expand.
DEDUPE is off and certain directories are GZIP9 (Which hammers the CPU)
Once solaris is up and running its great. ZFS vol sharing isn’t as good as I would of hoped though.

Note on this motherboard if your getting ZFS errors you may need to slow down your RAM in BIOS to improve dual channel RAM support.

FreeNAS is great but boy is it slow….

Anthony August 9, 2011 at 8:24 pm

Hi Logan you have a few conflicting statements.
“I’m not going to use a RAID-5 configuration and loose 1 drive, because I will have use a cloud back-up.”

“I chose ZFS because reliability is very important… will have all my family’s important stuff.” (Really?)

ZFS Has reliability only when losing drives capacity. RAID 0 is bad news and when it fails you *will* have major corruption issues. ZFS cant just rebuild your data without some sort of duplication.
ZFS Stripes to improve speed.
You will have ZFS panics and the potential for data loss is real (noone likes to recover from backup). The only difference is ZFS will notify you on a pool scrub.

The cost of one disk is cheep compared to recovery… Have you recovered 2+TB from the cloud before?

Logan August 10, 2011 at 12:33 am

@ Anthony

No I have not tried to recover 2TB from Cloud before, however I have tried to rebuild a RAID setup (drobo) when 1 drive goes bad.

I was thinking building 2 of these RAID setups….both identical… could I tell ZFS that one is a mirror?

Logan August 10, 2011 at 12:36 am

@ Anthony

Your first comment completely lost me about slow RAM, dedup, GZIP 9, and Stripes. I’m not a system admin…so i need a more simpler solution.

Logan August 10, 2011 at 12:37 am

Regarding ZFS, I’m most certainly going to use Ten’s Complement Z-410.

But I’m still open to other ideas… is TowerRAID good hardware? Or there better options?

Logan August 10, 2011 at 12:39 am

and what is “pool scrub?”

Erik August 10, 2011 at 12:46 am

I’m in agreement with most of the people that some kind of ZFS based solution is the way to go, especially if you want to ensure data integrity. Currently, I’d say that the best bang for your buck these days is the HP N36L Microserver. It’s small, inexpensive, quiet and hold 4 SATA disks off a SAS controller. AFAIK, it even supports the newer >2 Tb drives. I put together an N36L system for 680€ with 4 2Tb drives and 8Gb of RAM. For the OS, Nexenta, Solaris Express 11 with Napp-IT or FreeNAS, depending on the level of access to the OS you need/want.

I strongly recommend using RAIDZ for media files, but if you’re storing lots of little files, I would consider configuring two mirrors for better IOPS, but once you’re into the high capacity SATA drives, IOPS are going to be pretty lame anyway. Based on 3Tb drives, you’re looking at 9Tb of usable space in the 4 slots, although it may be possible to use the space for the DVD drive for a fifth drive which would bring you to 12Tb of usable, RAID protected data.

When it comes to the data redundancy, you can buy two, and replicate the data via zfs send/recv which is flexible enough that you can do your initial load locally with the two machines on the same network, and then move the copy to another location and continue the transfers over ssh to ensure a complete, viable offsite backup.

For an upgrade path you can either consider replacing the drives as newer ones becomes cheaper, or use the 16x PCIe slot to add a SAS card for external storage.

For sharing, it all depends on the client OS. For OS X clients, I find that NFS is rock solid, coupled with netatalk for Time Machine backups. zvols over iSCSI and SMB to Windows clients, depending on whether you need to share between multiple machines or not.

Damir Lukic August 10, 2011 at 4:31 am

RAID5/6 read speed should be the same as RAID0 read speed. Performance issue comes with write speeds, but if you have enough RAM on your RAID controller card, you should be OK with that. I suggest going with 3Ware or LSI RAID controllers.

RAID0 on multi-TB drives is a nonsense, as you cannot back it up in normal time, and you’re completely unprotected. You, of all people, should know that RAID rebuild process of just one failed multi-TB drive can take up to 2-3 days, sometimes even longer if using non-intelligent RAID controller. 🙂

Note: even space-saving snapshots (TimeMachine) won’t save you in this case as they do not clone the data on other set of drives.

nate August 10, 2011 at 8:31 am

My own setup was not geared towards being as cheap as possible, but after reading this I went and looked at the rebuild time of my local 4x2TB RAID 1+0 array when I had a disk failure about a couple of months ago. The rebuild took about 3 hours, and then the 3ware controller did an integrity scan of the array and that took about another 3.5 hours.

I decided to build a common system last year that acted as a storage system, a media system, as well as a general purpose server/workstation.

More recently bought a 1U server with 4x2TB SAS(also RAID 1+0) disks with ESXi on it and put it in a local data center here(on site support included) for my “cloud backup”(it runs many other things too). I get “unlimited” bandwidth at this particular facility, and I have the option to get up to 100Mbps of downstream (5Mbps up) from my cable provider for a good price (~$150/mo) though 500GB/mo limit ($0.25/GB extra for overages).

The cost for all this stuff was in the ~$7k range – though it does a lot more than basic storage. The server+colo pays for itself pretty quick vs the cloud for me – if I used the cloud I’d be looking at $350-400/mo, cloud storage alone from most of the providers I looked at would be easily $200/mo for 3TB of storage. So – cheap is relative of course.

With my own server co-lo’d I can fire up as many server instances as I want to do whatever I want, I virtualized the network on ESXi so there are two different networks on the main host, the virtual servers reside behind an OpenBSD firewall(everything runs on 1 system), so I can use the private IP space to run up more stuff if I want.

I agree refurb is a good way to go as well. The HP xw9400 workstation I bought for home use about a year ago was around $1600 w/3 years of support, vs the list price on their web site for the exact same system (new) was $4k. I upgraded the ram, video card, disks, added the RAID card, added a meager SSD for booting.

Also make sure to have a good UPS, I recently picked up a 1000VA Cyberpower Smart App Online (true sinewave) UPS with IP-based management card. Going to get an external battery pack for it(supports up to 4) to increase my run time. I was really happy to come across this UPS model, I had not used a Cyberpower UPS in years because they didn’t have true sine wave(when I did have them they worked fine except for things that needed true Sine wave). This not only had the battery expansion options but also the IP-management card for half or less than half of APC.
I recently retired a pair of 10-year old APC Smart UPSs when I moved to another state.

I expect the ~3.6TB of usable space I have will be enough for the next 3-5 years for myself.

And I’m sticking to hardware RAID with battery backed write-back cache(because SATA is slow) for the foreseeable future. I’ve been using 3ware on SATA and originally PATA for about 10 years now and have no complaints. I think I’ve had 1, maybe 2 cards go bad on me (both were more or less “DOA”) in the ~250 I have used.

stu August 10, 2011 at 8:39 am

For NAS storage (network accessible storage) appropriate for a home or small business — this site does a great job of benchmarking and reviewing all the various vendors.

Taylor August 10, 2011 at 10:32 am

Logan – get a Drobo. Let it do RAID. Back things up to the “Cloud” too.

Then, feel free to read up on RAID topics etc. It’s pretty clear that you’re not understanding it correctly (you don’t want RAID because you “lose a drive”, but you want ZFS because it is “reliable” – this is not an insult, but you have much to learn on the subject). Once you understand more, when it comes time to upgrade, perhaps you can use a more comprehensive solution. Until then, the Drobo is perfect for you.

Blake Irvin August 10, 2011 at 10:33 am

Nexenta offers up to 18 TB with their appliance software for no charge:

Nik August 10, 2011 at 11:58 am

The current version of freenas is great for this. I am using my old desktop that has an AMD64 X2 + 4GB of RAM. Over gigabit I can hit about 80MB/s write speed over CIFS(sustained) so it is not too slow.

Anthony August 10, 2011 at 11:56 pm

Pool Scrub is where ZFS will check for corruption and rebuild the data if it can.
If your going to learn one thing learn ZFS… It works on BSD, Solaris and MACOS

Important stuff needs RAID
ZFS Rebuilds the data only (not like a RAID Card)

Implementation arguments usually involve buy the hardware that someone has prepared for you (Costs alot)

Or design and build it yourself (Requires research So a cost of time)

Rick T August 11, 2011 at 12:06 pm

Logan, 5x3TB drives set up as a RAID 5 volume will net you just under 11 TiB of usable formatted drive space. The biggest issue with RAID 5 and multi-TB drives is the possibility that an uncorrected read error while rebuilding the array after a drive failure could take down the whole volume. If you can get up to hardware that supports RAID6 (or any other double parity solution) you will be in better shape. Mirroring RAID5 pairs may work but now you have even more disk space overhead (2n+2 instead of n+2).

One other number: a 1 Mbps link can transfer about 9 GB/day of data at 100% utilization, so copying many TB of data to the cloud for backup will take weeks to complete unless you have a very fast up-link at home.

Anon August 11, 2011 at 3:49 pm

I don’t know what version of zfs you used, but as of 2 years ago (when I switched to zfs), you certainly can add drives to a zfs pool. The downside is you can’t remove them. But you can replace them with bigger drives. So when yo want to upgrade, just swap drives.

martin August 12, 2011 at 8:34 am

I had similar challenge about 8 months ago. I needed storage for 10TB+ archive.

I ended up choosing 5-bay Synology DS1010+ (newer analog is DS1511+) NAS with 4x3TB drives RAID5 (1 slot free for future expansion).

Also, you can extend DS1511+ with 2 external e-sata DX510 5-bay enclosures.

It’s running linux and has decent web interface for management. Transfer speed over iscsi to windows server is as good as NAS boxes over 1GBit ethernet go (~100MB/sec) .

So far, it has been configure and forget type of setup, which I like.
It cost about $900 for NAS + drives.

Jacob Marley August 12, 2011 at 8:38 pm


Your not worried about an unrecoverable read error during a rebuild?

matt August 13, 2011 at 9:25 am

stay AWAY!! from ZFS unless you are a sysadmin type or SUN customer. I’m on zfs-discuss and the sheer number of problems people run into for simple stuff is mind-boggling. Linux software raid just works and LVM etc makes volume management trivial. See BackBlaze for component ideas. But personally I don’t use sh*t products so I use Areca 4gb cache controllers, Chenbro or Intel SAS expanders and four 4x internal 4-drive containers in a Coolmaster Centurion case. For casual home use, the 4 or 5-drive SATA-multiplier enclosure ( is just fine. There is absolutely no reason to run Raid5/6/Z on 1TB or bigger drives. The rebuilds take an eternity and the likelihood of loosing another drive is MUCH HIGHer than you realize. Raid1/10 is the way to go.

If going the PM (port-multiplier route) get 2 and put the mirror pairs across them. PM works well up to 3 constantly active drives per upstream cable. Total cost was < $350+ drives but the RAID controller was $1000 (had from another project). Some places to start:

Jacob Marley August 13, 2011 at 1:33 pm


I can understand RAID5 with larger than 1TB drive being a bad idea but why is RAID6 (two sets of parity providing two drives for redundancy) a bad idea?

If the rebuild times are a concern, why not split each drive into multiple softwareRAID partitions and create multiple RAID6 sets?

LX August 14, 2011 at 5:53 pm

It is still a research project in my lab, but can serve well the needs discussed here. Currently it only runs on Linux. Take a look at

It is easy to see you can replace a $1 usb hub with a more power USB switch or any other connector, and USB drive with a real hard drive. As for reliability, you can configure from RAID 0 to RAID 6, or even higher, say tolerate 3 or 4 disk failures, and of course you can do n-way mirroring, all via a very simple text-based configuration file. Once mount, it is just a regular file system/volume. If people have interests, we may add some GUI for simple use. Enjoy.

Dave August 15, 2011 at 12:14 am

“The folks asking this question aren’t full time sysadmins..”

Soon as I read this I thought “Get a NAS”. My most recent NAS (after replacing my wonderful ReadyNAS NV+ that was 4 years old) is a QNAP 659 Pro+ with 6x 2 TB drives in a RAID-6 configuration. 7.2 TB usable storage. More if you go with RAID-5.

Amazingly easy to set up (even for me) and I don’t have to think about getting geeky with it. Gives me time to get all geeky with my other endeavours (VMware, Hyper-V, etc).

Bought the unit and drives separately – just check the HCL (for whatever vendor you go with) and save yourself the hassle. It’s great to get geeky but I’d rather not mess with my data.

Steve R August 15, 2011 at 5:29 pm

I’d be curious to see some solutions to this problem that don’t involve a bunch of drives spinning continuously, 24×7. For a home system, who needs 24 hrs a day – IRL people sleep, leave the house, etc. Keeping a server going constantly is a horrific waste. Even when being used, it’s mostly for movies – so for 90 minutes of watching a movie (200GB?) the entire 10TB array has to be sucking electricity?

What I’d love to see as a solution to this kind of problem would be some sort of SATA card that could power cheap consumer drives on & off as needed. Sort of like the old hierarchical storage schemes that used robotic tape libraries, except using off-the-shelf SATA drives instead. Yeah, it might take 30 seconds to spin up the one drive you need with the movie on it, but who cares about the 30 seconds for a 90-minute movie that you only access once or twice a day?

Something like this isn’t a bank or an airline reservation system – I wish there were a more appropriate technology for this kind of situation than a 24×7 fully available server.

matt August 15, 2011 at 11:23 pm

that would be worse. a single disk failure would now affect 2 raid sets each rebuilding at different rates and causing all kinds of seeks back and forth. Yes, Raid6 buys your more time before complete disaster but even that can be too big a window. Linux software raid1 (maybe 10 too?) lets you do more than just a single data copy. Namely I can have a 3-disk raid1 or in other words 3 copies of the data so any failed drive means I have one more to go before I enter degraded mode. Raid5 etc. made sense back when people used 146GB or smaller 10k/15k drives so they could get some actual space out of all the itty-bitty devices. With the big drives especially, raid-10 means only one pair of drives is performance impacted while the rest run at full speed so the aggregate performance hit is pretty minor. A parity-RAID rebuild causes a MAJOR hit. For home use, it doesn’t matter.

ZFS RaidZ is a retarded design. They split the <128kb block across N-spindles. So to read just one block you have to involve all disks. In traditional RAID the historical convention is 64kb per disk and a full stripe is N-spindle multiplier of that. There are significant gains to be had when using a 256k-1MB even under random workloads. Head seeks is what costs the most time and the incremental platter read time of 256k or 1mb is practically zero compared to reading just 64kb.

matt August 15, 2011 at 11:32 pm

selective powerdown would be nice and the big boys (emc/netapp/datadirect/nexsan etc.) have this. But it requires an intelligent controller that can issue the necessary SCSI mode commands to slow the motor, or stop the motor, or even put the electronics in trickle mode. I expect Linux block layer will have this capability soon enough. Consumer drives don’t like lots of power state changes so that has to be considered as well.

Even a dozen drives doesn’t amount to much draw compared to corps with several hundred to thousands that might rarely be accessed and thus could benefit significantly from being spun down. Some of the LSI/Areca controllers will have this capability supposedly. With a big enough NVRAM or SSD write staging, the controller can leave spinning rust powered down until a threshold is reached.

Steve August 16, 2011 at 9:51 am

Good point about the scale. So how about this as a proposal – I’d like to have cloud storage available with a cheaper rate, in return for me being willing to wait for 30 seconds or whatever for me to begin accessing a file. After all, as a home user I’ll only be pulling down these big files once or twice a day, and realistically some days will go by when I never pull a file. If this mythical cloud storage company has a ton of customers like me, they can have 95% of their disks spun down at any given time – this should save a lot on power costs.

SR August 16, 2011 at 2:07 pm

@matt — latest zfs build puts metadata for raidz on a single drive.

mdadm +xfs+raid1 works quite well for my case, as some have pointed out.

allanster August 16, 2011 at 4:34 pm

Yes Matt, ZFS is nothing but trouble and is retarded, that’s why the hadron collider team chose it to manage their 15 petabytes of data, that’s why the large scale study to test ZFS proved it’s integrity and it passed with flying colors, that’s why a company with million dollar storage systems felt threatened enough to sue the creators of it.


Jacob Marley August 16, 2011 at 7:27 pm


Unless I misunderstood your reply to me, I think your understanding of RAID6 in general and Linux RAID1 across 3 drives is different than mine and possibly others 🙂
RAID6 generates two sets of parity for each stripe and each parity is written to a different disk.
If one disk has failed and is being rebuilt, the 2nd parity can be used to recover from an unrecoverable read error on another disk.
Also, during rebuild, the seek issues are near identical to what it would be during a RAID 5 rebuild.
Please note, for RAID1 across 3 disks, you only have two copies of the data (not three).

Finally, if I recall various ZFS video lectures, as far as reads are concerned, when reading blocks that are very close to each other (i.e. not adjacent), a sequential read of all blocks in the range is done.

martin August 18, 2011 at 11:26 am


I forgot to mention that we also have 2x10TB taperobots for backup.

I guess if I didnt have the taperobots I would buy another similar NAS box and configure it for daily sync or incremental backup

Anonymous August 18, 2011 at 6:08 pm

For some folks, unRAID might be a possible solution.

Snorre August 25, 2011 at 4:27 am

I have repurposed my old desktop (6-port mATX mobo with an Intel C2Q and 8GB RAM) with 6x 2TB Seagate Barracuda green-disks (5900RPM) running FreeNAS8 from a USB stick. The disks are set up as a 5+1 raidz pool, giving 10TB (9.1TiB) of available storage.

The reason for building this server is that my 4x2TB 2nd generation Drobo is complaining of low free space… *sigh*

Dave August 28, 2011 at 3:56 am

I’m always amazed by the way everyone just assumes ZFS is more stable than anything else and more reliable in the long term. And it’s all based on marketing fluff.

A word to the wise: before you go trusting all your data to ZFS, look up “ZFS Data Loss” on google, and follow the links. You’ll end up on Sun’s ZFS storage forum, which has a good amount of people complaining about catastrophic data loss with ZFS drives, and a complete lack of fsck-like repair tools for the file system. Most disturbing is the flip way the Sun (now oracle) support people and engineers deny, deny, deny, blame the user, try to weazle out of ZFS being responsible, and are otherwise quite flip about the whole situation.

As someone who has lost data on ZFS myself, I’d just say: Be careful.

Raoul August 28, 2011 at 8:30 pm

Forget RAID as you know it.

The future of storage (whether Enterprise or SOHO or even laptops!) is to have a checksum capable filesystem.

Think about it – as disks grow in volume, there’s statistically a greater chance of data corruption: Simple as that. (let alone silent data corruption that can occur now without you even knowing about it with RAID until it’s too late!)

No one likes to lose that photo that holds so many memories for them, and no company can afford to have staff twiddling their thumbs due to down time.

ZFS was the first file system that I know of that was available to anyone, and offers block checksums: And I’m sure it won’t be the last.

Here’s another situation to ponder.
Looking at life expectancies these days, I’ll probably live to be over 80 years of age no probs, so how many times do you think I’ll need to “transfer” data from an old system to a new system? A fair view times no doubt.
When you factor how many times your “treasured” data is going to be copied to another volume, I know I’ll never ever use anything else other than a checksum capable filesystem. – It’s a no-brainer.

I moved all my treasured data onto ZFS back in 2007 (MacZFS and Solaris) and it’s been an absolute relief to know that the data are protected far more than any current RAID system can offer. I’m currently beta testing the new ZFS offering by Tens Complement and it too will revolutionise the way consumers manage data, the same way Sun’s Solaris did for Sys Admins.


Anonymous September 4, 2011 at 1:02 am

Why not consider the use of ECC RAM along with a checksum capable file system?

Ok so the original question made the point “Price is important. Performance isn’t. ”

I don’t like to spend a lot of money and would rather re-purpose a system or build it myself. How do I ensure my data is as robust as possible at the sub-$500 price point?

I am already reading of people posting both for an against ZFS and BTRFS is still a long way away from being proven.

I don’t think a spendy RAID controller is the answer for a home user.

Jacob Marley September 4, 2011 at 12:18 pm


In the old days (Intel Pentium II/ AMD original Athlon) you could use non-ECC or ECC RAM with the same motherboard and CPU. Since then, (Intel first but eventually AMD followed) desktop/consumer CPUs/chipsets stopped supporting ECC RAM.

Now, if you want ECC, pay the premium for the entry level (single socket) server CPU and chipset.

Most of the consumer/SOHO NAS vendors don’t have ECC either, although some offer ECC RAM on their multi-thousand dollar business NAS models.

Back to your original question, why not run memtest on your box once in a while to make sure your the non-ECC RAM is in working order?

Michael September 5, 2011 at 7:45 am

@Steve R
I have a ZFS raid with raidz2 (similar to raid-6) and have never looked back. I am mostly worried about data corruption, here are research papers on data corruption vs ZFS, vs NTFS, vs hw-raid:

I heard of one guy that has put a power on/off switch on the cables to his disks, and when he is not using the raid it is turned off. He also has a 3TB disk as a temporary cache. Whenever he is done working with the data and the 3TB disk is full, he reboots and switches on the raid and copies data back and forth. Then he immediately switches off the raid. So, the raid is switched on only a few hours every week. Rest of the time he is working on the 3TB disk.

Avoid raidz1 (raid-5) because if you have 2TB disks or larger, and if one crashes, repair times can take several days. That puts additional stress on the other disks, and if one of them shows read errors you have lost your entire raid array. With large disks, use raidz2 (raid-6).

Anton September 19, 2011 at 11:51 pm

Re: ECC memory. Note that already mentioned HP microsever supports ECC memory. Here’s $300 solution that no consumer NAS can match.

And, you can install Snow Leopard and maczfs on it.

Jim Sherhart September 20, 2011 at 3:50 pm

Wow. Another excellent discussion and great to see Drobo mentioned by more than a few folks. What we like to say here at Drobo is the following:

1. You can likely build something that is less expensive and faster than Drobo.
2. If this is something you are seriously considering doing, you are probably not a target Drobo customer.
3. If this is something you know how to do but you value your time, then you are probably a target Drobo customer.

There seems to be a common perception that only people that can’t spell RAID buy Drobo. While it is certainly true that Drobo makes RAID easy enough for a 12 year old (, the fact is that most customers are quite storage savvy and simply value their time.

John (other John) October 5, 2011 at 2:47 pm

Very silly question:

if the stated idea is just getting movies off disk, can we assume not lots of them all the same time?

what i mean is, what is wrong with just high grade SATA spindles, an a mirror copy, same spec? Things like WE4 do not even power up unless needed, or you force that in the OS. And suddenly, for <$1K, you have 10TB, twice, and none it hardly ever being used, which puts MTBF into the stratosphere. No funky controller drivers, never a rebuild in sight, no OS configuration hell.

My point is that i do not know why this use required more than 100MB/s. Does any *movie* format use that much? (sure there are some, but this is no longer talking about "movies" but about 4:4:4 production data. Which case you go get some heavy lift kit.) Keep your indexes on a fast drive, back them up, and good to go.

I think there's more bunk in the low end of storage, as there is in a big name sales pitch. I'm confused how the word performance got used without description.

all best – j

John (other John) October 5, 2011 at 2:51 pm

and getting even close to 100MB/s is going to be expensive with NAS type gear. Respect to Jim above, because i think the question was not well put, i’m not diving in on that one! – j

Leave a Comment

Previous post:

Next post: