back to article Microsoft reinvents Massive Arrays of Idle Disks for Azure, 'cos IBM tape ain't enough

Microsoft is working on a new class of disk-based storage appliance for its Azure cloud service, because tape is proving too hard and too slow to use in some situations. Azure chief technology officer Mark Russinovich today told an event in Sydney, Australia, that Microsoft has deployed tape libraries to power its Azure …

  1. Hans 1
    Joke

    Is Azure not the prime example of a MAID or Is there somebody using their service ?

    1. TheVogon

      "Russinovich also said Microsoft has experimented with on-site natural gas generators for some Azure data centres. The generators feed a bank of batteries that power the facilities"

      They also power racks directly with gas:

      https://blogs.microsoft.com/firehose/2017/09/27/microsoft-partners-on-worlds-first-gas-powered-data-center-in-seattle-neighborhood/

  2. returnofthemus

    So much for Tape being dead...

    Just when you thought that everything in the Cloud was stored on Flash and now instead of focusing on Project Hololens Microsoft wants to disrupt the Tape storage market to try and save a bit of money for storing peoples garbage. Looks to me as if Microsoft is exposing itself and The Empire is Striking Back!

  3. Anonymous Coward
    Anonymous Coward

    Mark Russinovich should focus on SysInternals tools

    Mark Russinovich should focus on his SysInternals tools and his WinInternals book series. Since his joined MS nothing worthwhile surfaced.

    I love his ProcessExplorer, Autoruns, etc tools. Though, Mark Russinovich is not Dave Cutler (of VMS and WinNT fame). Mark, your old tools SysInternals need some love and regressions get more and more.

    If only there would be a ProcessExplorer-like tool for Linux. Gnome 3 system monitor is a joke (and ugly) and as bad as taskmgr.exe.

    1. Anonymous Coward
      Linux

      Re: Mark Russinovich should focus on SysInternals tools

      "If only there would be a ProcessExplorer-like tool for Linux."

      qps, opensnoop, dstat, htop

      1. Anonymous Coward
        Anonymous Coward

        Re: Mark Russinovich should focus on SysInternals tools

        "If only there would be a ProcessExplorer-like tool for Linux.

        qps, opensnoop, dstat, htop"

        How is this an adequate answer? Those are more console based "system monitor" alternatives than a real alternative to ProcessExplorer (which is a lot more feature-rich and has a graphical user interface). Sure, nowadays it's probably better to use an APM anyway.

  4. Loud Speaker

    Speed?

    As someone who has been a tape user for over 40 years, I do not recall ever selecting tape "because it is fast".

    However, IME, the life of hard disks which are constantly spun up and down will be short, nasty and brutish. And the power consumption during spin up is huge, so there may not be a net saving anyway. (I had an early example of RAID with 5 1/4" drives that claimed to take a kilo-amp for a microsecond during spin up. I doubt the power supply could source that, but it seemed to work. RFI was a problem though.

    1. katrinab Silver badge

      Re: Speed?

      How many times can you spin up a tape before it disintegrates? Manufacturers typically recommend you replace it after about 200 uses.

      1. returnofthemus

        How many times can you spin up a tape before it disintegrates?

        I take it they longer teach HSM in Storage 101 classes

        How many times can you spin up a hard drive before it fails and how many times can you write to flash before it wears out?

        1. Paul Hargreaves

          Re: Get burned?

          > How many times can you spin up a hard drive before it fails

          Depends on the drive. Modern drives can cope quite well - e.g. 600,000 on/offs: https://www.seagate.com/files/www-content/datasheets/pdfs/exos-5e8DS1954-2-1712GB-en_GB.pdf

          > how many times can you write to flash before it wears out

          Depends on the drive but wear levelling and the 'archive' nature of the workloads means that I'd be surprised if many fail. If placed into RAID sets of a reasonable number of drives you can multiply up the write endurance by all the drives (again, assuming a reasonable filesystem above) and you'll then not be able to sustain enough writes before your system is out of warranty since the controller will bottleneck.

      2. HobartTas

        Re: Speed?

        No, because "uses" as you put it is 200 times to fill up a tape which for LTO-5 according to Wikipedia here https://en.wikipedia.org/wiki/Linear_Tape-Open requires "80 end to end passes to fill up a tape" and so your ACTUAL TOTAL "Expected tape durability, end-to-end passes" is 16000 (80 times 200) and not the 200 you quoted.

        However, since the total size is 1.5TB this implies you could read 18.75 GB for one pass (1.5 TB / 80) so say for the entire data access operation you want to read just one (and only one) file (e.g. a movie) even up to this 18.75 GB size it could conceivably be contained on one pass but realistically most likely would be split over two and so you could quite comfortably do this procedure about 8000 times. Anything much smaller would most likely be readable with one pass only and of course I'm presuming the data on the tapes would be contiguous but that would be a reasonable assumption to make as the backup software would most likely be writing the files sequentially. Naturally I'd make a further assumption that each of the 4 bands and the 20 wraps per band get roughly the same amount of access because otherwise yes each individual wrap might only be good for 200 passes before the tape is worn out in that spot and the drive/software offlines the cartridge permanently due to "too many hardware errors" or whatever.

        Tape drives themselves have a MTBF of 250K - 1M hours or so and their tape load/unload cycles are also huge so I don't see them failing early for this reason either, but who knows the tape holding the header pin might snap off after say 1000? load/unload cycles but since its reinforced around that area then 8000-16000 cycles of single file accesses <= 18.75GB in size might still be quite reasonable for one tape cartridge.

        That's my 2 cents worth but someone more knowledgeable who actually uses LTO extensively may have a different opinion.

    2. Lusty

      Re: Speed?

      "However, IME, the life of hard disks which are constantly spun up and down will be short, nasty and brutish"

      I'd imagine that the global cloud provider, Microsoft, has more data on what does and doesn't hurt a disk than you do so they've probably thought about all that in quite some depth. Google did a study a while ago and concluded that none of those things hurt disks (heat, vibration, cycles, etc.) and it's down to luck and bad batches how long they last. That's partly why firmware updates so often trigger SMART alerts - they see a bad batch and proactively fail the drives.

    3. HobartTas

      Re: Speed?

      Yes, but modern hard drives that use 5-10 watts need around 20-25 watts to spin up for only a short period of time and most of that extra is the 12 volt feed so some reasonably large capacitors on that line could also assist and this only applies to consumer drives and setups whereas server grade gear such as SAS drives and raid cards/HBA's have things like PIUS (power up in standby) where to avoid such power surges the raid card spins up the drives one at a time, also if they do start up a batch of say 10-20 drives in PIUS mode you may only need a couple in that batch to actually start up and work and the others can stay in powered standby and also be mechanically idle until you shut that batch down and so that will save on head loading/unloading cycles for the unused drives.

    4. This post has been deleted by its author

  5. Anonymous Coward
    Boffin

    Thinking ahead

    Surely the precious source code for Window 10 should be stored on punched-cards to survive the EMP waves during WWIII?

    The few survivors will welcome having the seed banks buried in a Norwegian mountain and the code for Windows 10 to help them rebuild civilisation.

  6. BoomHauer

    Will the archival nature of tape ever make it obsolete though? Isn't there still demand for a simple, locked snapshot in time? The worry is that if it's on disk and on-the-ready, can it be altered? We all hate tape for the reliability and other assorted PITA reasons, but it's great for evidentiary reasons. Ask the bobbies.. http://www.bbc.com/news/magazine-20556330

    1. HobartTas

      Yes, cassette tapes are good for this reason as firstly they are analogue and the special machines they should be using also record alongside the audio track some special tone(s) so that if you cut and paste the tape or edit out sections of it this is detectable, but to answer your question if you use the ZFS filesystem you can snapshot the data on the hard drive so for example ransomware thinks its encrypting your data but all its doing is writing fresh data whereas the source data is in a read-only state and immutable, also if you record to LTO tapes you can also get a special variant WORM (write once read many) cartridge that could conceivable satisfy this requirement.

  7. HmmmYes

    Oh fuck this. Its been solved zfs.

    Just take it, write a native bsd block layyer

POST COMMENT House rules

Not a member of The Register? Create a new account here.

  • Enter your comment

  • Add an icon

Anonymous cowards cannot choose their icon

Other stories you might like