Build your own 180TB NAS for $US1,942.59 (plus disk)

Cloud storage company Backblaze may not quite have the cachet of the folks playing in the Open Compute Project, but that hasn't stopped the company open-sourcing the design it's cooked up for the JBOD-like rigs it uses to power its service. And after receiving a million page impressions for the release of its first design, the …

[Silverburn]

(plus disk)

Ah, so what you mean is your 180TB NAS actually costs nearer £12,000 (based on retail prices of 3.5" 4TB drives).

Still...that's not bad. No data on redundancy method though.

[TheVogon]

Re: (plus disk)

I make it nearer £10K assuming £170 a disk. Plus you shoudl be able to get a discount for buying 45 at a pop.

[EvilGav 1]

Re: (plus disk)

Given the cost for 3TB drives is hovering around £100, i'd say it's not worth building a 4TB rig - adding 45B of additional storage increases the cost by around £3,500, which just doesn't make sense economically.

Should be able to build the chasis for ~£1,500, so the 180TB version is around £9,000 or 135TB for around £6,000.

[Matt Bryant]

Re: EvilGav 1 Re: (plus disk)

"Given the cost for 3TB drives....." Given rebuild times with 3TB disks, and the benefits of using more spindles for "hot" data, I'd suggest a tiering of smaller disks (say 600GB) for the tier 1 "hot" data, then the large disks for archives, etc.

[Yet Another Anonymous coward]

Re: (plus disk)

>No data on redundancy method though.

They run raid 6 (2 parity drives) on each box and the data is replicated over at least one other box.

The server part to distribute customers data as efficenctly as possible accross different hardware, is their secret sauce

[DJ Smiley]

Re: (plus disk)

Theres no reason you can't build a raid on it...

[Danny 14]

you're gonna need some nice cooling to keep a 4U 180TB box running smoothly.

[Ben Norris]

Not really, remember that this is based more towards archiving with most of the drives sitting idle in power saving mode. It isn't meant to be a like for like replacement for a SAN

[DJ Smiley]

Not really.... thats the entire point of their design.

[Kebabbert]

"Preserving the data"? Data corruption!

The backblaze company are using normal Linux filesystems, is it XFS?, and as research has shown, those filesystems are susceptible to data corruption. There is always a very small risk of data corruption, but if you have very large amounts of data, you will surely get data corruption. And 180TB is bound to have data corruption. The more data, the more problems.

Research about Linux filesystems being unsafe:

http://www.zdnet.com/blog/storage/how-microsoft-puts-your-data-at-risk/169

Large amounts of data always face data corruption, says Amazon engineer:

http://perspectives.mvdirona.com/2012/02/26/ObservationsOnErrorsCorrectionsTrustOfDependentSystems.aspx

Research shows that ZFS is indeed safe, and protects your data:

http://research.cs.wisc.edu/wind/Publications/zfs-corruption-fast10.pdf

[Matt Bryant]

Re: Kebbabfart

".....Research shows that ZFS is indeed safe....." Can it be clustered yet? No? Oh, so when your one ZFS server dies you lose access to all your data. FAIL! This has been pointed out to you countless numbers of times, Kebby, so please stop with the ZFS cheerleader routine until it is enterprise-ready.

[TheVogon]

Re: Kebbabfart

You could could build an enterprise supported and clustered NFS 4.1 / SMB3 solution to do this on Windows Server 2012 - use the iSCSI server on each storage array, and then a clustered head end that mounts the volumes in a resilient RAID config. That way you could also use deduplication.

[Allison Park]

thanks Matt

I didn't know ZFS couldn't be clustered. Crazy we dont use it as we prefer storage from EMC/IBM/Netapp

[Matt Bryant]

Re: thanks Matt

"I didn't know ZFS couldn't be clustered....." ZFS is such a control freak bit of software, insisting on direct access right down to individual disks, that it doesn't even play well with hardware RAID cards.