back to article Pure Storage to punt out supersized FlashArray system

Pure Storage’s Accelerate event in San Francisco earlier this month previewed a coming high-end FlashArray//m system with more than three times the usable capacity of the current range-topping m70. The new system will arrive later this year and give Pure a five box line-up, unless one of the current//m products is dropped. A …

  1. Anonymous Coward
    Anonymous Coward

    what's the point?

    Lets just pretend for a few seconds that more than .01% of the IT community actually needed 1.5PB of latency so low they needed an AFA of this size...why not just scale out?

    Can you just imagine what the cost of an M90 is likely to be? TCO or not, thats a hell of a lump of <1ms latency than most workloads just don't need.

    1. Nate Amsden

      Re: what's the point?

      Pure doesn't do scale out does it ? Or do you mean operate multiple independent silos of storage ?

      e.g. with the 3PAR 20k series you can start with 2 controllers, scale up to 8 controllers, and scale out to 4 arrays with their federation tech (up to roughly 16PB of SSD storage and 14TB of write cache(or mix in some spinning rust if you like too, 3PAR doesn't care, the 20k goes to 6PB raw capacity with disks or run hybrid, it's so flexible!), I'm sure that will go up more once the 8TB SSDs come out for them).

      Having a petabyte of storage behind what I assume is 2 controllers for Pure(active/passive still?), doesn't seem like a solution I'd want to risk using. I'm sure they have some good products, just not something I'm interested in using.

    2. Yahtzee24

      Re: what's the point?

      I believe the cost per GB will make it worth it to put all of your data on the FlashBlade. If a company needs a PB of capacity, why have 2 arrays just because it doesn't feel right to put some stuff on an AFA? If they have more than 1.5 PB, then maybe it makes sense to get a cheaper, slower solution to support the Pure, instead of a 2nd Pure.

  2. Pavlov's obedient mutt

    NetApp AFF on that list?

    this retina display thingy on this 'ere macbook must be dodgy, I cant see the Netapp AFF on that list..

  3. Vaughn Stewart

    Let's correct the math

    Disclaimer: Pure Storage employee

    Chris I'd like to correct the math behind the capacity. In the post you overlooked the 20 flash modules that sit behind the bezel of the controller chassis. Thus a FlashArray with two shelves is comprised of 68 flash models. With each at 8TB in size, one would end up with 544TBs of raw capacity for 1.5PBs of usable capacity. The formula for raw to effective usable applies a 5:1 data reduction ratio (which is lower than the average across our install base) to the remaining raw capacity following the allocation for RAID-3D and FlashCare.

    -- Cheers

    V

  4. Chris Mellor 1

    Replying to Vaughn Stewart of Pure

    Thanks Vaughn. That's me taken out to the woodshed - as I'm bush league :-)

    So there are 20 drives in the coming top-end FlashArray//m-whatever controller enclosure. That would be extra 20 x 8TB = 160TB totalling up to (384 + 160) = 544TB, which makers 1.l5PB usable more achievable, as you say. Lovely system. Also the chart then needs updating.... sigh .... as the new FlashArray would be ahead of the VNXe3200 (500TB) but still behind Kaminario's K2 and its 740TB.

    1. Vaughn Stewart

      Re: Replying to Vaughn Stewart of Pure

      Nice quip mate ;)

      The math form raw to effective usable should apply equally to all FlashArray //m configurations and models.

      - cheers

      v

  5. Afrojazz

    Have you ever tried to check "analyst" information before publishing it?

    In spite of it's very silly to compare modern AFA by raw capacity, please tell me how "Mr. Analyst" managed to get only 380TB of raw capacity for SF9605/9010 if every node has 10 960GB SSD and there will be 100 nodes?

    And as it was mentioned by commenter before, where is NetApp AFF?!

    It can has tremendous 22.12PB raw capacity for NAS only cluster and 7.37PB raw for unified cluster.

    That is simply capacity monster, not HPE.

  6. Anonymous Coward
    Anonymous Coward

    Nimble's AFA?

    I wonder how Nimble's new AFA would compare? If I'm reading the specs on their website correctly, it looks like a full 4x AF9000 configuration (fully scaled-out and up I guess) already has about 1.5PB usable capacity.

    I don't see a figure for how many VMs are supported. But I wonder how this config would stack up against this Pure box, especially in terms of pricing?

    1. NickT

      Re: Nimble's AFA?

      Disclaimer: Nick Triantos Nimble Employee

      Nimble's AFA scales to 553TB raw per system, x 4 for scale-out = 2.2PB Raw. Effective would be 8.1PB using a 5:1 data reduction.

      How many VMs it supports? Well a vague question deserves a vague answer :-)

      It depends on what the VMs do.

      This applies to everyone except those who have figured out a way to beat common sense. :-)

      Cheers

  7. Anonymous Coward
    Anonymous Coward

    high performance!

    500,000 IOPS out of 1.5 PB is only 333 IOPS per TB.

    Not quite what a lot of people might be expecting from an all flash solution - it's more like SAS disk in terms of IOPS density.

    Before anyone flames me, I know IOPS isn't the only measure of performance, but it's hardly the normal flash message which is that it's so fast "you're paying for the capacity - the IOPS are free".

  8. HLC

    Capacity is only effective if you can buy it

    Disclaimer - NetApp marketing guy

    I have to admit - maximum capacity isn't the all-flash attribute I normally spend my day on. But here are the numbers for the NetApp AFF product which are not easily ignored (well they shouldn't be):

    22PB Raw Capacity = 480 drives * 3.8TB = 1.8PB / Array * 12 Arrays / Cluster = 21.9PB / Cluster

    18PB Usable Capacity = 82% or 1.5PB / Array or 17.9PB / Cluster

    80PB Effective Capacity @ 4:1 ~ 80PB / Cluster

    Most flash experts are turning their attention to scale-out. It's one of the reasons we are so bullish on SolidFire which adds a new storage controller for every 10 SSDs.

    And finally, at the risk of stating the obvious, the capacity of FlashArray is effectively zero since it's not shipping ;-)

    1. Zerolab

      Re: Capacity is only effective if you can buy it

      You are maybe the only one turning your attention to block scale-out. Solidfire OPEX and overall TCO does not follow the cost curve customer and the industry expect from a flash array, this is way they are mostly irrelevant and will remain so. Regarding Netapp there is not much to say, just look at the company earnings product revenue in constant decline and service and maintenance revenu going up, signs of a dead company. Talking about TCO, just look at Pure 45% overhead and architecture claims, buy 100% HW but use just 50% and maybe get 5:1 or at least until they are force to change their dedup algorithm due to patent infringement....

      1. Anonymous Coward
        Anonymous Coward

        Re: Capacity is only effective if you can buy it

        We dont see HP rocking it with their flat revenue trend either sport so I wouldnt get too cocky.

  9. Reydrx

    Missing Elements

    The analysis mentions TCO; seemingly as an afterthought, but never mentions the U's or power consumption associated to the arrays listed in the analysis.

    Scaling out has numerous cost-related factors, most of which were overlooked. Why? The information to complete a thorough comparison can be found in most whitepapers, so why leave those factors out while giving TCO an "honorable mention" nod?

  10. Yahtzee24

    m10 to "m90"

    "We assume that the upgradability from the entry-level m10 to the new system, which we'll call the m90 for convenience, is preserved with Pure's Evergreen scheme."

    I suppose I could be wrong, but I'm pretty sure that there is no way to go from an m10 to the new FlashBlade. The chassis are completely different.

    1. Anonymous Coward
      Anonymous Coward

      Re: m10 to "m90"

      You could upgrade M10 to M20 etc. but that would not be Evergreen as that just covers 'equivalent' controller not a step up to the model above. Flashblade it a totally different product so no upgrade path.

  11. This post has been deleted by its author

  12. Anonymous Coward
    Anonymous Coward

    $1,000 per TB

    If that is true, that is a really good deal. The market prices for the comparable Flash arrays, including Pure, is about $5,000.

    1. Anonymous Coward
      Anonymous Coward

      Re: $1,000 per TB

      I'd be careful on that math by marketing. As everyone knows (several blogs on this) the usable capacity for pure drop as you use capacity, and to recover from failed drives. Yes, the Meta-data takes away from the available capacity as the usage gets higher. So day one at 1% usage maybe $1000 per TB, but you're going to lose up to 45% usable capacity when you fill it up, which is going to piss a lot of users off once deployed.

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