back to article Top dog EMC: Clear off, hybrid upstarts, VNX2 is a million-IOPS monster

EMC is developing a follow-on VNX storage array that will combine flash and disk drive storage in a hybrid arrangement and deliver a million IOPS of performance. It will run a Rockies update of the VNX operating environment software, Flare and Dart. This information comes from accounts of EMC World presentations earlier this …

COMMENTS

This topic is closed for new posts.
  1. adam1942

    double capacity

    Double capacity from 4 petabytes to 16 petabytes. Shouldn't that be from 4 to 8pb?

    Will be interesting to see how many SSD's you require to get 1mil OPs as some consumer SSD's (standard format not PCI-e) are rated for up to 85,000 - the key words being rated and up to. Will also be nice to see what size of SSD's they offer and if they can make the price any more competitive than the current pricing for the SSD's used.

  2. Nate Amsden

    lame

    This stupid race with so many different folks trying to produce million IOPS systems is terrible.

    EMC blasts SPC-1 for for not having a realistic workload and then brags about a million iops number that is no doubt assuming a 100% read workload with a low I/O size.

    There have been two different self proclaimed "million IOPS" systems put to the SPC-1 test that I know of so far, one fell short by 60% the other fell short by 40%. Both systems were 100% flash.

    I bet EMC's would fall short by about 80% given the hybrid design.

    Show me a million IOPS system that does it under 100% write and I'll be impressed (my own VM workload is 96% write). Oh and no cheating with RAID 0 or anything!

    1. Matt Bryant Silver badge
      Thumb Up

      Re: lame

      ".....no doubt assuming a 100% read workload with a low I/O size....." Indeed. I tell our application people to not talk about IOPS but to talk about application traffic requirments, it's better for sizing. Playing the sizing IOPS game, you can make a quite poor array configuration look great by assuming 100% reads (i.e., static data that never has to be upated) and cache hits of 50% (anything even close to 30% is a miracle in the real World!). And usually they are talking about the fast disk in the system, ignoring the slow disks also attached to the same controllers that are used for archiving or backups, and which will often be asking for cache and controller bandwidth at the same time as the fast disk in real operations. The vendors know this, any IOPS "guarantee" will have a page of caveats in tiny print.

      Worst case, when your application people really don't know what they want, ask the vendor to do the sizing with 50% reads and 50% writes and a maximum of 10% cache hits, then get what they specify in to test with real data. Never, EVER, buy on an IOPS figure.

    2. Anonymous Coward
      Anonymous Coward

      Re: lame

      I'd like to see the current VNX doing 200,000 IOPS, never mind one million IOPS in a new VNX, It's to be hoped they've informed Intel of this requirement, as it seems like the rest of the mid range crowd, they're reliant on Intel to provide performance kickers these days. Saying it will overlap low end VMAX is quite amusing. If it could really do 1 million IOPs outside of a lab environment, then from a performance perspective VMAX would be dead in the water. This is just the usual marketing bluster for the non initiated, EMC seem really fond of leaking these pre announcements the last couple of years. I remember Chuck Hollis accusing Netapp of doing this, He referred to it as raiding the road map. I suppose if there's not much on the horizon, then you need to generate some kind of buzz to maintain momentum.

      1. Anonymous Coward
        Anonymous Coward

        Re: lame

        Doh...I just worked out where the 200,000 IOPS number came from. VNX 7500 supports 1000 disks, If all of those are 15,000 RPM disks then that's 200 IOPS per disk, assuming no writes and therefore no raid overhead, It obviously doesn't take into account realities such as vault drives or hot spares which would be sat idle, but anyway here goes.

        1000 disks @ 200 IOPS = 200,000 IOPS :-)

        So we now know the 200,000 IOPS number is complete fiction and wouldn't be achievable outside a EMC lab never mind SPC-1 or any thing remotely resembling a real world environment.

        1. Anonymous Coward
          Anonymous Coward

          Re: lame

          It's not that simple, a 15k disk is rated for about 170 IOPS worst case (assumes random read).

          The drives can do far far more than 170 IOPS given the right conditions and these are the conditions the vendors love to give.

          Regardless, the drives are not the barrier in hitting 1m IOPS in a VNX. The CPUs are.

          1. Anonymous Coward
            Anonymous Coward

            Re: lame

            I was taking the piss :-)

        2. Anonymous Coward
          Anonymous Coward

          Re: lame

          If all of the drives were SSD drives that would mean about 5k IOPS per drive and 5 million IOPS

  3. FeST
    Devil

    blurring messages

    Everyone who knows realities behind scenes aware that EMC is pumping interesting messages to market, unconcious-pure minds, storage buyers. If EMC would like to show its self confidence, must make a real show against rivals in front of them- in a gladiator arena- SPC benchmarks.

    1. Anonymous Coward
      Anonymous Coward

      Re: blurring messages

      Never hath dumber words been spoken. SPC is a waste of space and I was upset when EMC started posting again - SPC should have been left to die. If people bought based on the SPC ratings then I'm pretty sure *they* would be the unconcious-pure minds and out of a job. Pure IOPS benchmarks are only a small part of the buying decision.

      EMC technology is very simple to understand. It's clear why they are number 1.

      They don't excel at anything technology wise, but like all vendors they say they do (DMX / VMAX the exception). They take products that are merely good (not the best) and sell to the business.

      Yes your new niche solid state juju array may beat EMC hands down but there is a lot of comfort in going with the herd and buying from the larger players.

      1. Anonymous Coward
        Anonymous Coward

        Re: blurring messages

        When did they start posting SPC results ? I think you probably meant to use "we" rather than "they".

        Say what you like about SPC but if EMC had a result, we wouldn't have to keep listening to these marketing numbers.

        1. Anonymous Coward
          Anonymous Coward

          Re: blurring messages

          No, definitely 'they'.

          I saw a deck about a year ago where they were touting figures from Isilon and VNX.

          Upon deeper digging it appears this is the SPEC benchmark. Point stands though, these benchmarks achieve nothing but a bit of willy waving.

          1. Anonymous Coward
            Anonymous Coward

            Re: blurring messages

            Yeah but at least SPC has a cost per IOp which severely limits the vendors ability to build silly solutions (excluding some of the crazier IBM configurations). They either have to build a reasonable solution i.e not hundreds of SSD's and multiple arrays strapped together or discount the hell out of it, at which point you have them over a bit of a barrel.

      2. Anonymous Coward
        Anonymous Coward

        Re: blurring messages

        Of course , a SPC result cannot be only and definitive decision criteria.Whenever a new result presented or in need of comparing them, checking all configuration details, performance metrics are pretty enough for all.Frankly, I'm not getting serious about $-gig-perf relationship of SPC results, because it can be manipulated in any way.

        But, you may accept or not , SPC gives technically undertandable- reasonable evaluation criteria, proofs What you should be looking for benchmarking and whatever you would like to see in it.

        EMC hasn't started to post results hence you don't need to worry about others and keep yourself paralyzed under EMC marketing stuff.

        1. Anonymous Coward
          Anonymous Coward

          Re: blurring messages

          Only a vendor would see value in SPC. As a user they are a waste of time and it’s nothing to do with EMC marketing. Amazing how everything is an ‘EMC’ conspiracy and EMC brainwashing (we are a HDS shop for Tier 1 and Netapp for Tier 2). It could never simply be the case that a real customer would find these amazing results so useless!

          It does not give you an understandable or reasonable evaluation of a product in any sense. It’s a load of guff that only vendor engineering departments understand or care about. It gives me the maximum IOPS in a tweaked product for a configuration that does not resemble my environment and a test that does not resemble any typical environment workload. The storage world is very different to the supercomputing world where I only care about MIPS and that is the only thing.

          So what does it come down to? Whoever can get the most drives and the most CPUs into one architecture. I could have told you they would be the fastest without any benchmarking. All it takes is a vendor with scale-out architecture to keep adding more and more pieces of their hardware and they keep the top spot. They let someone beat them for a little while and then all they do is re-run the test with another load of heads, nodes, or storage processors to retake the top spot and gain some free advertising as well as another WOW powerpoint slide.

          Can you hear that sound? It’s the sound of many vendors waving their willies around and around in a game between themselves. Most sensible users take no notice beyond a cursory glance, but quite a few people not only take note but also try to catch some of these willies in their mouth.

  4. Matthew Morris

    28GB/s

    That is almost something - considering that clients who are running heterogenous environments that need 28GB/s of bandwidth today...can buy 4 6616s from Violin (32GB/s of bandwidth) they don't have to wait for any software releases from EMC. Heck they can go ahead and install the other 8 6616s in the single 42u Rack and provide 48GB/s of bandwidth to customers who need it today.

    And thanks to Cisco - we have already proven our ability to deliver high performance for VMware and to HP, Cisco our ability to deliver that performance for Oracle, MS SQL Server database applications as well.

    No EMC Software Licenses, No EMC Support Costs, No Oracle Exadata Software License/ Support Costs, No additional Oracle Database License increase for the newer X3 gear....

    Just plane ole simple, fast, reliable, highly available storage....(for density switch out the 6616 for the 6232s..)

  5. M. B.
    Pint

    Guys guys guys!

    "Our new stuff is faster than our old stuff by THIS HUGE AMOUNT!" - Every storage vendor in the history of data storage

    Drop the SPC garbage. Drop the marketing fluff. Storage guys should not be getting excited over statements like this, we've been duped and misled too many times before.

    Show us REAL WORLD mixed workload performance with low latency and the ability to deal with huge IO spikes and give us real pricing information with real data resiliency factored in and the licensed features people actually need to make the thing do what it says on the box. Then show us how you do it better than everyone else. THEN maybe I'll care about the marketing spew. Also, tell your fanboys to shut the hell up.

    Sorry for the rant, this is not EMC-specific but they're certainly a convenient target right about now. Ask me how my mid-range storage RFI is going. Long day.

    Beer because of beer.

This topic is closed for new posts.

Other stories you might like