Is there a shingled disk in your future?

Is there a shingled disk in your future?

Summary: Disk vendors are running out of options when it comes to increasing disk drive density. One increasingly attractive option: Shingled Magnetic Recording (SMR). What is it and will it succeed?

SHARE:
TOPICS: Storage
10

What is shingled magnetic recording?

In today's disk drives tracks are separated by a small space or gap. In SMR drives, this gap is removed to achieve greater density.

Write heads lay down a wider track than read heads need. In an SMR disk the write tracks are overlapped, leaving narrow tracks that are fine for the read head, but which cannot be easily overwritten without destroying the data on one or more adjacent tracks.

In practice, the tracks on an SMR disk would be laid down in groups or bands of tracks. The band enables a partial rewrite of the disk. How big that rewrite is depends on the size of the band.

Why is SMR attractive?

Seagate is researching SMR because the major alternatives, Heat-Assisted Magnetic Recording (HAMR) and Bit-patterned Media (BPM) look to require massive investments that drive vendors - hit by the drop-off in PC sales - may not be able to afford (for more on HAMR and BPM, see Engineering the 10TB notebook drive). SMR drives can be built using today's production lines and technology - with 1 significant exception.

The SMR question mark

The downside of SMR is the fact that you can't rewrite data in place: you have to rewrite an entire band of tracks. However, this is similar to the problem that NAND flash controller developers have been grappling with the last 5 years: you have to rewrite an entire 256k block (or greater) of flash, not just a 4k page. 

As data is deleted and capacity freed up, the remaining good data has to be collected and rewritten so unused capacity can be reclaimed, a process known at garbage collection. Managing that garbage collection and other processes requires the same kind of virtualization of the disk blocks flash controllers have been performing on SSDs.

But another option is changing the file system to treat an SMR disk more like a tape device: gathering blocks to be written together and then writing all of them at the same time. Seagate has supported academic research into this option at Carnegie-Mellon University.

The Storage Bits take

Drive vendors Seagate, WD and Toshiba have been hit with a double-whammy: SSDs have devastated their high-margin enterprise drive business; and tablets and smartphones are stunting their bread-and-butter PC drive business. They've finally moved into the higher-margin add-on storage business, but I expect anemic financial results for the next several years. 

If they don't preserve their $/GB cost advantage over SSDs they'll lose even more volume. SMR looks to be the easiest way to keep growing areal density, but they'll need much more sophisticated drive controllers or file systems to make it work. 

The controller model will be the fastest to implement, but at higher product cost. New file systems would mean lower costs for drives, but they are years away, if ever. 

The disk drive business has never been easy. It's looking even tougher going forward.

Comments welcome, of course. Hybrid drives may offer the easiest path to SMR, since there is a controller and plenty of flash already on the drive. But do you want a 10TB notebook drive?

Topic: Storage

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

Talkback

10 comments
Log in or register to join the discussion
  • Another factor that hits the HDD industry...

    is cloud storage. Some who used to store TB of data on their SAN storage are now sync it out to Dropbox, SugarSynx, SkyDrive, or other cloude storage.
    gamoniac
    • Cloud hits their margins. . . .

      But the data is still stored on a drive, but one purchased on a bulk buy at lower margins.

      Robin
      R Harris
      • NUH UH!

        It's in the cloud, dummy!

        The cloud is a magical substance made up of ponies and fairy dust =D
        Tea.Rollins
        • Ponies & Fairy Dust?

          I was told that all the data was all stored at the end of the rainbow in the crock next to the gold.

          or at least it was something to do with a crock...
          bigroj
  • Are you sure about the block write thing?

    Normally, with NAND, you only have to erase the page and not the entire block of the chip. But then again, there may be some higher level aspects on some of the controllers that don't do that. I always live on the software/interface layer right next to the NAND and I am 100% sure you do not have to read/erase/write an entire block. In short, the chips support doing page access data only with block erase supported when doing big bulk writes (great for really big files like a movie); the higher level SSD controllers, however, may only support block read/erase/write.

    Can you clarify?
    Bruizer
    • Block erase is the general case

      Bruizer, while I've seen NAND spec sheets that say you can erase 3 pages of a block without rewriting the entire block, that doesn't seem to be the general case. Due to either garbage collection or multi-page updates - which even if written to a different location leads to garbage collection - full block writes are the general case.

      I'm curious what NAND you are using. Is it on a SOC or microcontroller? Because consumer flash, which is what most SSDs are made with, has the block program erase requirement.

      Robin
      R Harris
      • Mostly industrial piece parts.

        From Atmel, Micron and Samsung. I have never had the advantage of working with a SSD controller but get to manage the individual read/erase/write based off of actually physical memory addresses in the chip by talking directly to the chip. While page sizes have grown (from 64 bytes to 256 bytes to now 4K and 8K) and blocks are now huge, we have always been able to erase and write individual pages and use the block commands only for large bulk operations (like resets, production initialization and the like).
        Bruizer
  • Back to the future of disk recording?

    In the old IBM mainframe drives, from System/360 (1964) on, most applications on most operating systems treated each of the many tracks like very short loops of magnetic tape: create data at the end of the track, erasing the track as you go, in whatever block format your logic required. Only specific applications rewrote records within tracks (preformatted fixed block direct access files and, later, VSAM keyed or direct access fixed-record files). The PC/MAC file systems, which only format a drive ONCE, were a great simplification, removing the link between logical file format and physical allocation. Now they want to go BACK?
    jallan32
    • The old 8" HP drives were a bit better.

      I had an old HP system for about 3 years (I used to always use obsolete hardware for fun) with a full sized HP Digitizer (The digitizer was pretty cool). The system had a 2 Meg 8" drive where all files had to be contiguous and the system did not provide a de-frag (free space, not files, would become very fragmented). As an exercise, I wrote a simply de-frag. It would typically take about 2-3 hours to run (data running over the old IEEE-488 bus) but would really free up lots of usable space.

      Ahhh... Those were the days.
      Bruizer
      • Lots of fun...

        It is like you had a lot of fun back in the days like I did... It's amazing how the technology moves in cycles, and sometimes in circles, redefining itself by using old paradigms and bringing news ways to do things more efficiently.
        When you look at a Samsung Note tablet with 32Gb and you double to 64Gb just adding a thirty bucks SD card and compare the total cost of the toys we had before, it is really funny. I still remember my 'first huge' hard drive with 5 MB... with a cost of more than the Samsung Note I mentioned...
        Even though I miss those challenges we had back in the days, I love this time we are living and all the fun playing with toys... I mean, working... This week I installed a DB2 Express-C into an old AMD with Linux Mint (supposedly not ported...). Digging to it, an apt-get found me the libraries and installed them closing all dependencies... and now I have the playground I wanted... See, lots of fun. I wish the day could have more than 24 hours... or could slow down a little bit. Cheers.
        FuzzyIce