Buy a 180TB array for 6¢/GB

Buy a 180TB array for 6¢/GB

Summary: After five years, the folks at Backblaze are back with a completely redesigned Storage Pod. The V4 is faster, better AND cheaper. And open source, so you can build your own. Here's what you need to know.

SHARE:

Backblaze is an all-you-can-eat backup service for $5/mo. As such, they couldn't afford expensive brand name RAID arrays, SANs, or NAS boxes, so they rolled their own.

They now have over 100,000TB of their original Storage Pods in versions 1, 2, and 3. All had 45 drives using nine 5-drive SATA port multipliers - not the most reliable design - which meant five drives barfed when a PM went down.

4x performance
The new design has two HighPoint Rocket 750 cards, each delivering 40 6Gb/s SATA 3 interfaces through 10 mini-SAS ports. Much faster and less contention because each drive gets its own dedicated SATA and power port.

The Rocket 750s mount on an 8x PCIe 2.0 bus, giving a 4x performance boost over the earlier design. This enables much faster RAID syncs bringing up a new Pod and should - though Backblaze didn't speak to this - also enable faster rebuilds of replacement drives.

The direct connections remove the port multipliers, simplify the wiring and isolate connector issues - a common problem - to a single drive. Each mini-SAS cable fans out to 4 SATA connectors. Like this:

R750_picture
Graphic courtesy HighPoint

 

The old Pods had 2 power supplies for capacity, not redundancy, and that was expensive. The new Pod has a single high-efficiency power supply that powers everything and should be more reliable as well as cheaper. If you were using just a few Pods, you'd mirror data across a pair for availability.

There are other changes. New side rails improve cab mounting and access. Boot drives are on the rear for easier wiring and access.

5¢ a gig
But Backblaze's price for a 180TB Pod is now a record low $9305, including drives - just over 5¢ a gigabyte of raw capacity. You'll pay 6¢/GB if you buy from their supplier - 45 Drives - but they also offer optional redundant power and boot drives.

Backblaze has published the complete parts list so you can buy the chassis from 45 Drives and bolt the parts together yourself and save some more money. Its an open source design so you can even bend your own metal if you like.

The Storage Bits take
The Storage Pods are designed for Backblaze's particular need: low-cost storage for online backup. They would choke and die on a call-center transactional workload.

But as we gather more data - backups, log files, web content, video - we typically access it less. And while SSDs are great for transactional workloads, disk drives are still competitive for large sequential reads and writes and are way cheaper.

As drive prices drop - 5 and 6TB drives are starting to come on the market in volume - Storage Pod 4 costs will drop as well. If you need lots of capacity there's no lower cost solution.

Comments welcome, as always. Say, with 6TB drives that would be 270TB per Pod and a petabyte for about $50k next year. I remember when 500MB was $50k! 

Read the entire Backblaze v4 blog post here.

Topics: Storage, Hardware, Open Source

Kick off your day with ZDNet's daily email newsletter. It's the freshest tech news and opinion, served hot. Get it.

Talkback

11 comments
Log in or register to join the discussion
  • Wow

    That would be so much fun working on that project. Cool stuff... I do not need 180 TB, but boy that would be fun to have.
    schultzycom
  • Yea... What would you do with that much space?

    I have a few TB on my main desktop at home and I still don't know what I would do with it honestly.
    Jimster480
    • 180 TB Application

      I have a client who has 550 TB of videos. So they would need 3 180TB systems. I remember when I thought my 10 MB drive was huge but times and data requirements change.
      jdb@...
      • Yeah, this is backup media for multimedia

        As a photographer, I'm constantly looking for large, cheap backup. Add video, and the need increases exponentially. I laugh when I read cloud backup articles and people brag about cobbling together free space from Dropbox, Drive, etc. This is good pricing for big needs.
        big red one
      • Another solution

        Maybe you should suggest to your client that he should invest in a good Bluray burner and some bluray disks, just to be on the safe side, mind you.

        Personally, if I had that much video, they would all be copied, with redundancy twice over, to optical media of one form or another.
        bart001fr
      • Yes photogs

        If you are photographer even as a hobby, you would need at least 5 Teras, but for a movie maker, 1 Gig is only seconds of film in 2K or 4K RAW, one movie would take at minimum 5 teras.
        orendon
    • Not for the average joe

      This is for commercial applications, not backing up a pc.
      Brock Jones
  • Less will be more

    In the future, all info will be centrally located. New data will be compared to existing data and only the deltas will be added to the database. For instance there will be no need to store the human genome, just the chimpanzee genome plus the slight differences in the human genome. Certainly a separate copy of genomes for every individual working in comparative genetics will be a thing of the past.

    This will become increasingly sophisticated. That picture of you in front of the Eiffel tower will be reduced to already known and recorded elements. Eiffel tower, biometrics of your face, image of GAP sweatshirt (catalogue # j8fk.008), weather conditions for that day, directional orientation to establish shadow pattern, what your hotel was serving for breakfast that morning, and what time of day you usually take a dump. In the unlikely event that anyone ever wants to see that picture, not the dump--you at the Eiffel tower, it can be assembled on the fly, including the expression on your face, to a 99.9% degree of probability.

    In fact, data is about truth and accuracy and in most cases, truth is what we say we like but actually don't. Thus there is no need to even take the picture in the first place and you can record your entire vacation simply by selecting a destination, available day trips, length of stay, your movie star companion, clothes from a designer store, and how much hair you want put back to fill in that bald spot on your head. That is, you would do all this if you are a hands-on person. For most people, the only new data is choice of destination. The system already knows when your vacation is and can deduce the rest of your likely choices by your already established buying patterns. So enjoy your staycation. No need to turn on the tv, the system already knows when you will sit down to watch the video re-enactment of the vacation you would have had if you had had one.

    My point, and I do have one, is that the notion that people will need more and more data storage is just wrong. Amazon has already figured this out and does not store your media on the cloud. They don't store a separate copy of Cloudy with a Chance of Meatballs 2 for each customer and soon they won’t have to store anything at all because they will be able to tell everybody that CWCM2 had none of the charm of the original, and based on what they know about you, you won't like it and you will believe them because your experience has been that they are always right.

    It gets worse. You think of yourself as a unique individual, but 99% of the people who bought Prada shoes at age 27, have dined more than twenty times at an IHOP, and had a parent who died before they were twelve love Barbara Streisand. It turns out that there are only about 17,000 personality types, each of which can be determined by analyzing fewer than 1000 personality indicator data points. Once your personality type is established, all the system needs to know is Type #11,648 plus any factors that will cause some fluctuating deviation from that norm.

    The email you write your boss about why you were late today can be deduced from existing data. In fact, she has already received it and deleted it. Her system-generated reply expressing annoyance at your use of passive voice is waiting to be read and deleted by you. When you get fired, HR will be able to reconstruct all the communication between you and your boss (who was type #07,463, with deltas 46, and g87), by knowing your type, the daily cafeteria menu and your team’s record of achieving company goals during the period of your employment. In fact, they can see the whole thing coming and your pink slip is already in your inbox. No worries, your better-match-employer has already been determined and that is why your office doesn't look familiar; you are already at your new job, which is staying at home and watching a video of yourself where you are at work, doing something that doesn't need to be done because a very limited set of data is all that is necessary to determine what you would do, and that work has been automatically accomplished on your behalf. Subsequently there was no need to record what “you” did because that can be reconstructed.

    Yes, the pride I had in the huge 30MB hard drive I got with my 286 computer 30 years ago was quite reasonable. By 2017 it will be large enough to store all the necessary data for a multi-national mega-corporation. So buying more storage now is like piling up rocks to throw. They aren't going to do you much good come WWIII, which will be decided by simulated battles based on existing troop position data. You would be the type to get your foot blown off, wouldn't you? Fortunately the products that you would buy with the disability check you would have received have already been delivered, stolen from your porch, thief apprehended based on his personality data indicators, returned to your porch, then picked up by the thrift store that you would send them to after enjoying them for a few days. You aren't the type to watch a pointless work simulation, so a simulation is cuing up now of you with one foot, hanging from your new super strength helicopter drone. Sit down, put both feet up and relax! It's a brave new world.
    Producto Endorsair
    • Awesome

      Please tell me that you have your own column or blog. I would love to follow.

      Uhm, more to the point, 5MB was a stupendously massive HDD back when I started in PC's (8086, without co-pro) and cost what the machine did the previous year.
      LaMorte
      • Give replay to the matrix, code #1234;254teuywt66762

        already in your experience ;)
        orendon
  • Thank you, Robin

    I've been following the Backblaze story ever since your first report about them some five(is that correct?) years ago.

    I was very interested to hear how they overcame the drive shortage two years ago. It's a very good story.

    This post and the Backblaze blog post should be a vrey good basis for anyone who wants to build their own server/storage solution without going to the expense of 180TB. After all, who individually needs this much capacity?

    But for the individual who needs 10 to 100 TB, a very roomy case with a lot of disk mounts could be the very thing. All they would then need is one of Highpoint's many offerings of the SATA controller cards and the rest of the computer (i.e. mobo, psu, ram, boot drive, etc...). This shouldn't end up costing an arm or a leg, since they can most effectively use an old computer which still works but is not the main one they work on everyday.

    I only scanned the Newegg offerings of Highpoint products and I am sure that they have something to suit just about everyone's needs.

    Thanks again for keeping up to date with the Backblaze saga.
    bart001fr