Intel Storage Platforms for the NAS Market

The higher end segment of the SOHO / SMB NAS market uses Intel's Core-series and Xeon CPUs to deliver the required performance while supporting a large number of drive bays (typically more than 8, in a rackmount form-factor). Those are not the subject of discussion in this article. Over the last four years or so, Intel's play in the other tiers has been mainly with its Atom CPU lineup. A little bit of history gives us some perspective.

History

In 2010, Intel launched the first Atom-based storage platform (codenamed 'Bandon'), combining a D410 or D510 processor with the 82801R I/O controller (belonging to the ICH9 family). It was capable of supporting up to 6 drive bays, and we even reviewed LaCie's offering (5big Storage Server) using this platform. The Atom D410 / D510 were both based on the Bonnell microarchitecture (coming under the Pineview codename). Later in the year, Intel also pushed out updated Atom CPUs (D425 / D525) with slightly higher base frequencies for use with the storage platform. The QNAP TS-659 Pro II was one of the Atom D525-sporting NAS units that we reviewed.

Pineview (45 nm) was followed by Cedarview (32 nm) in late 2011. It was still based on the Bonnell microarchitecture, but offered higher clock speeds and better performance per watt. Almost a year later, the official storage platform (codenamed 'Milstead') announcement with the Atom D2550 (followed by the D2700) and the 82801JB I/O controller (belonging to the ICH10 family) was made. The focus was on the improved graphics capabilities and HDMI output, which enabled interesting new use-cases related to media playback and surveillance video. LaCie's 5big NAS Pro and the Synology DS1812+ were some of the NAS units that we reviewed in this family.

Between the rollout of products based on the Bandon and Milstead platforms, the NAS market for home consumers and power users started experiencing rapid growth. This market demanded a different set of features compared to SMB requirements. NAS vendors had to provide extensive multimedia support (not in terms of playback on a display, but the ability to manage and stream content). Being a cost-sensitive market, vendors were loath to use a two-chip solution (main CPU + I/O controller hub). Towards serving this market, Intel decided to re-purpose their Berryville CE processor (a 32nm SoC with two Bonnell-based Atom cores as the host CPU) which originally targeted the STBs. For storage solutions, these CE 53xx chips were relaunched in their Evansport avatar and pitched as a media server platform. The CE 53xx SoCs also happened to have a H.264 encoder in addition to a multi-format decoder (giving it the capability to act as a transcoder). We have been pretty bullish on Evansport, having reviewed devices from all of Intel's announced partners (Asustor, Synology and Thecus).

Silvermont into the Picture

In moving from 32nm to 22nm, Intel completely revamped the microarchitecture for their Atom cores. Bonnell was replaced by Silvermont, bringing out of order execution and other improvements into the picture. With so many code names associated with Silvermont-based products, we thought it would be best to present a bulleted list indicating the markets which Intel hopes to address with each of them.

  • Bay Trail
    • Bay Trail-T: Atom Z36xx and Z37xx series for tablets
    • Bay Trail-M: Pentium and Celeron branding (N-series) for notebooks and AIOs
    • Bay Trail-D: Penitum and Celeron branding (J-series) for desktops
    • Bay Trail-I: Atom E38xx for the embedded market
  • Merrifield
    • Atom Z34xx: Low-end to mid-range smartphones
  • Moorefield
    • Atom Z35xx: Premium smartphones
  • Avoton
    • Atom C2xx0: Microservers and cloud storage
  • Rangeley
    • Atom C2xx8: Network and communication infrastructure

From our analysis of the various products, we believe the highlighted ones make sense for the NAS market. While the suitability of the Avoton and Rangeley for the NAS market is without question (they have a large number of SATA ports and PCIe lanes integrated into the SoC), the Bay Trail-D parts are quite interesting.

The various possible components in a Bay Trail SoC are given in the diagram below.

Depending on the target market (as specified in the bulleted list above), some of the components in the above block diagram are cut out. For example, Bay Trail-T does away with the SATA and PCIe lanes. Bay Trail-D is more interesting to us in this article, as the QNAP TS-x51's Celeron J1800 belongs to that family.

Introduction QNAP Goes Bay Trail
POST A COMMENT

51 Comments

View All Comments

  • ganeshts - Tuesday, July 1, 2014 - link

    Yes, I have done it and it works. Reply
  • mannyvel - Tuesday, July 1, 2014 - link

    How did your PC reconstruct the RAID? Did you have to match the RAID driver versions on your linux box/pc with your nas?

    Do you have a blog post/etc on your steps/process?
    Reply
  • ganeshts - Wednesday, July 2, 2014 - link

    Use UFS Explorer. I have a RAID-1 example here:

    http://www.anandtech.com/show/4510/lg-n2a2-nas-rev...

    UFS Explorer will show you the volume as-is and does the heavy lifting of RAID reconstruction. I am not sure what you mean by RAID driver version, as these are all just software RAID, nothing proprietary involved - standard EXT4 file system with mdadm.

    I will try to write a post on RAID-5 rebuild sometime in the near future.
    Reply
  • mannyvel - Wednesday, July 2, 2014 - link

    Oh, what you're saying is they're using the standard linux LVM-based md-raid, not a hardware raid implementation - which is why UFS Explorer works. Reply
  • GreenThumb - Monday, June 23, 2014 - link

    snakyjake>> 4) Upgradability. I can't afford to keep purchasing a new complete system. And what about new vendor software?

    good point - does QNAP charge for software updates to the same hardware?
    Reply
  • ganeshts - Monday, June 23, 2014 - link

    Nope.. They haven't charged till now (and none of their competitors in this market segment have that practice, either). Obviously, don't know about the future. Reply
  • bsd228 - Monday, June 23, 2014 - link

    These prosumer NAS units have made tremendous strides in the past 2 years, so I'll agree that for the disinterested, the DIY route makes less and less sense. But for those that care, it's as strong as ever. Haswell brought us 10W processors that can do everything Atom could, but 10x faster. For virtualization wishes, or background conversions of dvds/blurays to more highly compressed mp4s, this is essential, but with it sitting idle most of the time, 80W units were wasteful, hot, noisy.

    The key failing still present in this QNAP series is the lack of ECC memory. My HP Microservers have that that for 4 generations, as well as the ability to run numerous VMs (OS dependent - I use solaris). It was until the Gen8s stuck with a poor AMD processor and still isn't where I'd like it to be. So my next one will be truly DIY, not a nearly turnkey microserver.
    Reply
  • npz - Tuesday, June 24, 2014 - link

    It should be noted why open source projects and video enthusiasts have not exactly embraced Quick Sync. Aside from encoding quality (high bitrates is ok, low bitrates is bad), it's just not that flexible.

    The VAAPI table shows it's limited to h.264 High Profile @ Level 4.1 for both encode and decode.

    This means: 8-bit 1080p, max 30fps, with max of 4 reference frames.

    What if you've ripped your blu-rays for archiving at the best quality possible and now you want to transcode for mobile devices, or other home devices like roku, etc? Encoding is possible with QuickSync considering the constrained specs you're targeting, but DECODE is now impossible!
    Reply
  • ganeshts - Tuesday, June 24, 2014 - link

    With storage being so cheap nowadays, you could just back up your Blu-rays at the best quality (i.e, just remux to MKV, no re-encoding) -- and they would still be compliant for QuickSync decoding.

    Btw, I have seen Quick Sync decoder work great for 1080p60 clips (but that is on Windows using Eric Gur's code). To be frank, I have not seen any non-10-bit H.264 file that is not decodable via Eric's filter. (You could theoretically make a 120 fps clip with crazy high bit-rates that might choke QS, but those are not seen in real life). It even decodes the Planet Earth 16 ref. clip in hardware on the Intel Haswell NUC.

    It might be that VA-API just put that 'restriction' in, because if you want to claim full L5.1 support, that would require supporting really crazy encodes that no one probably does. We will have to see how it works after I get a review sample in hand.
    Reply
  • npz - Tuesday, June 24, 2014 - link

    I haven't tried the windows filter since I normally use MPlayer, VLC, ffmpeg, but that's interesting to know. It could be that the QS hardware supports greater than L4.1 but not L5+ The VA-API lists a 40 Mb/s limit, so that would be one limitation for L5.0

    It's true storage is cheaper, but not that cheap yet. Storage density is also an issue. Recently just built 40TB machine, and it probably won't be enough later. BD remuxes are absolutely out of the question (BTW some Japanese BDs even use several times the bitrate as US BDs!) Also, most of mine L5.0 or thereabouts (meaning res, bitrate, ref limitations match L5), but there is also the additional issue of using features that is not supported by high profile. If you don't specify a profile, and just use CRF for constant iamge quality + very slow preset + tune (which adjusts psychovisual parameters), then x264 will pick the most optimal and efficient settings. I've noticed constraining to high profile will increase the size of my encodes.
    Reply

Log in

Don't have an account? Sign up now