Comments Locked

15 Comments

Back to Article

  • DanNeely - Thursday, September 28, 2017 - link

    Using the momentum of the drive platter to power the nand writes in a power loss situation instead of a battery/capacitor scores points for clever design.

    Outside of 512b legacy support does the driver need to write to the flash in normal operation? From the description it sounds like its not needed in normal 4k operation. If that's the case, how much does the increase in writes to it affect endurance when supporting a legacy platform?
  • bill.rookard - Thursday, September 28, 2017 - link

    Remember, the NAND is only utilized during a power loss situation where the DRAM has data waiting for writing. Otherwise, the NAND just sits there doing nothing... whether the drive is emulating 512b sectors or not, the use of NAND is handled by the firmware when it detects:

    1) A power-loss condition
    2) uncommitted data in the DRAM buffer.
  • extide - Thursday, September 28, 2017 - link

    Where does it say that? It says it is a Persistent Write Cache, so a Write Cache that is persistent, sounds to me like it would be used all the time -- perhaps it reserves some for use in a power loss situation, but even if it only has 8GB of NAND there is only 256GB of DRAM so you could still use most of the NAND as a cache even if you reserved 256MB strictly for power loss.
  • Samus - Thursday, September 28, 2017 - link

    Reminds me of Toto's impeller generator for their auto flush solenoids. They have a tiny solar cell as well, but most of the energy required to charge the supercap and operate the solenoid is generated by the flow of water.

    Love seeing clever solutions to generate a bit of momentary power!
  • nathanddrews - Thursday, September 28, 2017 - link

    The headline says NAND cache, but the table says 256MB DRAM cache. Is this an error? Just curious how the NAND fits in here without it being a hybrid...
  • Kvaern1 - Thursday, September 28, 2017 - link

    As I understand it the only purpose of the NAND (PWT) is to store the content of the DRAM cache in case of a power failure.
  • nathanddrews - Thursday, September 28, 2017 - link

    Yeah, if that's if it works, then that's really cool.
  • bill.rookard - Thursday, September 28, 2017 - link

    From what I can gather through the article and logic'ing out what probably happens:

    The drive functions as a normal drive, with the DRAM cache to buffer incoming writes. The NAND sits idle. In the event of a power loss, the momentum of the spinning drive is used to maintain power to the DRAM cache and then powers up the NAND which it flushes the DRAM to until the stored writes are cleared out of DRAM.

    Then, when the drive powers up again, the controller will read the NAND for the not yet committed data, commit the data to the platter, flush the NAND and go idle and wait for data read/write requests.

    The nice thing is that this really does simplify board design, removes components like capacitors and such which take up space and can still be prone to failure, and utilizes the existing parts with few minor changes to the firmware and maybe an extra IC to siphon power from the motor during spin-down. Very clever.
  • ads295 - Thursday, September 28, 2017 - link

    You're saying the momentum of the spinning drive is enough to power both the DRAM and the NAND?
  • BrokenCrayons - Thursday, September 28, 2017 - link

    I certainly should be more than enough. We're talking about 256MB of DRAM that has to be written to NAND. That's two low power ICs and the drive's controller getting a second or so of power from the mometum of seven disk platters spinning down from 7200 RPM. We aren't talking about a DIMM with like eight RAM chips or a full SSD so the electrical demand is going to be quite small.
  • DanNeely - Thursday, September 28, 2017 - link

    They also don't need to do anything fancy in the flash with block management or wear leveling; just treat it as a 256MB buffer and do a sequential write. Probably they only thing they're doing is tracking bad blocks to avoid them.
  • jozevolf - Thursday, September 13, 2018 - link

    PROBABLY is the word. PROBLEM is, it simply doesn't work, at least not on this model. I have 10 of these drives (MG06ACA10TE) and using it in ZFS system. If I mark these drives as having non-volatile cache in my OS (OmniOS; /kernel/drv/sd.conf: sd-config-list = "ATA TOSHIBA MG06ACA1","cache-nonvolatile:true";) and power off the system by unplugging power cords during heavy writing, the ZFS gets corrupted.

    The drives also don't ignore cache flush command which they should if Toshiba engineers were confident enough in their Persistent Write Cache/Power Loss Protection technology.

    I've tried the same with ZFS on Linux. Pool of mirrored vdevs, consisting of only these drives and zfs_nocacheflush module parameter set. Same result: ZFS corrupted.

    Toshiba is also ignoring all my support requests regarding this problem. Is there anybody out there using these drives?
  • MrSpadge - Thursday, September 28, 2017 - link

    "If the drive is not helium-based, this slightly lower performance is explainable — it is harder for arms and heads to move in air environment (which has 7x higher density than helium), so “air” drives are a bit slower than helium-filled HDDs."

    Anton, apart from the density difference that's wrong. The sequential performance of HDDs depends solely on spindle speed and linear bit density. The helium filled drives are faster because they can use higher density platters. I think that's because there's less vibration due to friction.

    What you're saying would result in different access times. However, moving the heads quick enough is never a problem - just use a stronger motor. Moving it precisely enough is challenging. And the practical limit to access time of a HDD is the time it takes the platter to rotate to the proper position. Which helium filling doesn't change.
  • Anton Shilov - Friday, September 29, 2017 - link

    Thanks for bringing this to my attention, MrSpadge.

    In fact, I incorrectly read MiB/s vs. MB/s and then tried to explain it. Now that I've fixed everything, the drive looks to be on par with its 10 TB rivals when it comes to sequential data transfer rate.

    As for precise head positioning, HGST says that helium indeed helps with this as fluid flow forces affect head positioning. The problem is that HGST has used a different micro actuator with its He drives than it used for other drives, so we cannot really tell whether it's helium or helium + micro actuator or the micro actuator alone.
  • andychow - Sunday, October 1, 2017 - link

    The helium allows you to use more platters. A rotating disk in a fluid (air and helium are compressible fluids) creates laminar flow. Too much laminar flow becomes turbulent flow. With helium, you can have either more platters at the same speed, or higher speed with the same amount of platters, without introducing more turbulence i.e. vibrations.

Log in

Don't have an account? Sign up now