back to article Hey, who wants a 40TB all-flash Pure box? I dunno, you got $160k?

Pure Storage has given birth to two more all-flash arrays, and updated its storage software by adding data protection features. The FlashArray FA-400 line has had an FA-405 entry-level product and an FA-450 high-end model added to the existing FA-420, creating a three-model range. The FA-405 has up to 40TB of "usable capacity …

COMMENTS

This topic is closed for new posts.
  1. Anonymous Coward
    Anonymous Coward

    ... what?

    Uh... what? 40TB of 'usable capacity' that's "actually 2.75 to 11TB of raw flash".

    Are these boxes magical? Do the flash chips inside them randomly multiply, sorcerors apprentice style, before popping out of existence again if they get too close to the 11TB limit? Does nobody know the actual amount of flash inside? You taking a wild guess that it's somewhere between those limits?

    And how the hell can a 'pure flash' device have an 'usable capacity' four times the (apparently indeterminite) amount of flash it's stuffed with? If I cat from /dev/urandom, how much am I gonna actually write to it?

    Colour me utterly confused.

    1. Anonymous Coward
      Anonymous Coward

      Re: ... what?

      You must be getting hard facts confused with Pure's infamous marketing spiel.

      "maximum usable space of 250TB (34 to 70TB raw)"

      Hey they're just numbers, right? They don't have to add up. Quick, do something to distract everyone... Hey look, our CEO shaved his hair off.

    2. jai

      Re: ... what?

      The difference between usable and raw capacity is due to Pure's software deduplicating data.

      so presumably, you can get 40TB of data on there, but only if you're writing 14 identical copies of 2.75TB of data ?

    3. MityDK

      Re: ... what?

      They are talking about effective capacity, which takes into account their deduplication (which they do on a 512B block size, so it's extremely effective), their compression (which is done on variable block size starting from 512B, so again, super effective), and their pattern removal. Thin provisioning has nothing to do with it.

      Actual raw usable from the 5.5TB unit is about 4TB, because they use some space for their Purity OE and the Metadata tables and whatnot. The numbers they publish for RAW are just actual raw disk without any overhead applied to them, so the numbers are a little less. But their focus is on minimizing the data footprint rather than raw vs usable, they care about effective capacity.

      It has everything to do with data efficiency. Their target markets are Database, VDI and Virtualized environments. VDI will give you 10x+ effective capacity because of so much duplication of data. Virtualized environments are up in the 6x range.

      Database is not much for dedup but will compress ok. Dedup comes into play for database when people start making copies of the database, because Pure uses pointer based zero footprint copies that use just a bit of metadata. So Database can vary from 2x efficiency up to 8 to 10 or more depending on how many copies DBAs are making.

      Now, you would never want to put unstructured data on these arrays that doesn't dedup or compress... that would be very bad and your $/GB would be through the roof. All Flash arrays aren't for every kind of data type until prices come way way down and densities go way way up.

      Anyhoo--that's how they are doing it, and as someone who has installed a dozen of these things for various customers, it works fantastically well. Super fast with great efficiency rates, typically we see 4x-10x, DB on the low side, and VDI on the high side.

  2. Yesnomaybe

    I too...

    ...don't understand the numbers. Usable capacity is normally LOWER than raw capacity. Normally lose some 20% or so.

  3. Anonymous Coward
    Anonymous Coward

    numbers...

    They're basing their numbers off their 'ticker' dedupe savings data they get from existing customers and empirical data. I don't necessarily subscribe to selling arrays based on numbers of an 'it depends on your dataset' type of mentality, however they're not being cloak and dagger about it. They do share the evidence in plain sight and they're not hiding raw capacity numbers, so the buyer knows what they're getting into beforehand. As for the /dev/urandom comment, I think it's highly unlikely that a customer has 100% non-dedupeable data in their environment.

    1. Anonymous Coward
      Anonymous Coward

      Re: numbers...

      Go and have a look at the ticket on their website. Then look at the source... it's utterly fabricated.

      Or, as Pure might describe it, "marketing".

  4. Anonymous Coward
    Anonymous Coward

    You have heard of thin provisioning and dedupe haven't you guys?

    On our Pure array we have 8.2tb usable but have provisioned out over 300tb, however we have only used 3tb of space.

    I for one love the Pure array, it's in a different league to anything we used previously.

    Sure it's not all been smooth but I can't fault Pure Support for their abilities either.

    1. Anonymous Coward
      Anonymous Coward

      Re: You have heard of thin provisioning and dedupe haven't you guys?

      "however we have only used 3tb of space."

      But how much non zero uncompressed data have you actually written to use 3TB of raw space?

      nb - Thin Provisioning is very old news.

  5. Annihilator

    New marketing spiel

    First time I've ever seen someone reference "40TBs" instead of just "40TB"

  6. Pavlov's obedient mutt

    an IO is an IO right?

    So as claiming any number of IOPS without stating read or write, or random or sequential is about as useful as a chocolate teapot, how about some real world examples that we can use to compare apples to apples....

  7. M. B.

    I have seen...

    ...a situation where the system didn't nearly meet the marketing claims as far as the storage efficiency numbers were concerned. They were told 30x which got a good laugh out of everyone at first. After "data modelling" Pure said no less than 5.5x (not that they put it in writing though).

    Actual result was 2.3x and not enough capacity in the array as it was originally sized, so the customer had to buy something more cost-effective to round out the capacity deficit (they needed about double the storage efficiency number to ingest everything with a bit of breathing room).

    Unfortunate and very disappointing. My God is it fast, though! The CIO was pissed off and seeing red all they way up until all the comments on how amazing the ERP and CRM systems were running started to flow in.

    I do believe that Pure ended up upgrading them to newer shelves at some point for dirt cheap dollars though. Never heard how that turned out.

  8. M. B.

    Aha!

    Out of curiosity I asked my contact who went down the Pure route how the whole thing turned out since I heard the "cheap disks" thing second hand.

    Turns out Pure threw an agreement together whereby they would renew their support early and co-term the support for an additional shelf, and they'd get the shelf free because Pure screwed up. So they didn't replace anything but they did get a big spiff out of the deal.

    Might have had something to do with their "fill in" array being a Nimble though (they are also very happy with it, lucky bastards).

  9. EricB

    It's quite obvious all of the commentators on here have yet to play with a Pure storage array. Yes, you actually can squeeze 40TB of data on much less due to a three different deduplication passes and compression algorithms that are a part of the Purity OS. seriously go to the website and read about how you can scope 40TB of data onto a box that is spec'd to hold much less.

  10. Anonymous Coward
    Anonymous Coward

    So long as you have the right number of cores and memory bandwidth, I have no trouble believing it. I came up with some (un)real numbers here on my hand-built design. So when you control all of the parameters and are willing to test to destruction with a serious budget, yeah you get something like Pure, et al.

  11. Anonymous Coward
    Anonymous Coward

    It depends on your data!

    Like anything it depends on your data contents to what you actually get from dedupe and compression. For VDI and VMware these things are hard to beat. I have 24TB provisioned on ours using 2TB of actual space. Now as far as actual dedupe rates we are doing all VMs on ours a combination of VDI and VMware both windows and linux. RHEL 4/5/6 and Windows 2003,XP, 2008R2, and Windows 7... getting right at their quotes 6-1 data reduction. So that means we would physically be using about 12TB on most systems only using 2 on theirs. We liked them so much we ordered 2 more for some other stuff, and we are looking at another shelf or two for this one so that we can move our remaining really large VMs over and potentially even some of our smaller test databases. During our intitial testing we used a chassis of HS22Vs over 8Gb FC and could get 5.7GB/s using iometer pure read sequential, using 4Kb 100% random read we could hit 270-280,000 iops seems to have a bit of a write issue but still more than acceptable for us when doing 50/50 read writes hitting only about 900MB/s but their new OS claims to have made some changes there though we havent tested it yet.

    1. Tiluchi

      Re: It depends on your data!

      Hi!!

      Reading your comment is like honey to a bear, i'm just getting on with several VDI projects here in Bolivia, and already chose PureStorage among others, i'm convinced their technology is diruptive, will work with a mix of RHEV and VMWare clusters, also working with HS22Vs, and Dell M620 servers.

      Would you please share some more data of your environments?, looking forward to get the most from flash performance + vdi.

This topic is closed for new posts.