- Storage White Papers
There have been a few interesting articles discussing EMC’s anticipated release of Project X, also known as their XtremIO acquisition from earlier this year. Tom Isakovich, CEO of Nimbus Data has a blog that does a virtual teardown on the leaked hardware specifications. Robin Harris also chooses to raise the stakes with a commentary on EMC’s problem child. Somehow I seem to have become embroiled in the EMC mudslinging as my name is quoted as one of Robin’s timescale references.
However, putting the EMC name calling aside, what we’re starting to see is some leaked information (that could of course be fact or fiction) on how the XtremIO product will look. I agree with Tom’s appraisal that, compared to the rest of the market already out there, the new EMC offering doesn’t seem to be pushing the bounds of innovation. In fact, the technology as deployed seems to be simply more than a variation of their existing hardware technologies. Why? Let’s look at the details.
CLARiiON on Steroids
- Two storage controllers
- Two BBU (battery backup units), e.g. standby power supplies or SPS
- One DAE (disk shelf), 25x 2.5″ SSD
As the bricks scale out to form complete systems, Infiniband switches are used as the brick interconnect, however the other components scale out pretty much consistently.
Each brick seems to be built to the standard dual controller architecture – two controllers with volatile cache, requiring battery backup in order to flush data in the event of a power failure. The DAE even seems to be the same as ones used elsewhere. Pretty much a typical design, like CLARiiON for instance. The specifications for a single brick are:
- 250,000 random read I/O (4K)
- 100,000 random write I/O (4K)
- Power: 1150W
- Space: 6U
- Average latency: 0.5ms
By comparison, a single “brick” from Whiptail delivers 250,000 random write I/Os at 0.1ms latency using 200W and 2U. I could go on and choose more examples (and to be fair, some of the details, such as I/O size and mixed workloads need to be normalised for comparison), however I’m not seeing a revolution in what EMC are offering, but rather a late-to-market product that falls behind the best of the competition.
The Standard Counterargument
In anticipation of a vendor backlash, let me preempt some of the expected comments.
- Other vendor solutions aren’t as resilient as the EMC offering. Well, that one can be argued either way. Some vendors like SolidFire, are using many multiple nodes to provide distributed load balancing and resiliency, without the need for rackloads of batteries. Other vendors are building resiliency into their solutions by removing volatile cache into the controllers or moving it to the disk shelves. It would be naive indeed to think that these startup companies haven’t gone through due diligence on ensuring that their solutions provide the levels of resilience and availability customers will expect.
- It’s a 1.0 product, expect more from 2.0 and beyond. Unfortunately, many existing vendors, e.g. Violin Memory, are already on to their second or third generation of hardware and continue to develop and innovate. EMC would be required to invest a huge amount of investment effort to leapfrog what’s already out there in a mature market. Consider EMC’s VFCache as an example. This is an OEM card that still lags behind other existing vendors in the marketplace.
- It gives our customers choice. Actually I think it creates a confusing situation. An all-flash VNX or VMAX could potentially deliver similar levels of performance with extra functionality to boot. There becomes a difficult discussion for the sales team to have regarding which platform best suits the customer requirements, especially if those customers are already familiar with the VNX & VMAX product lines.
- Other companies as startups may not be around that long. That’s true – there is risk investing in using technology from a startup company. However if you are spending lots of money, this is part of the due diligence you perform and then it becomes a risk factor you weight against all the others when making a judgement. It doesn’t make startup bad companies; it merely means you have to decide on your level of risk aversion.
The Architect’s View
EMC’s flash strategy started strongly in 2009 when they were the first vendor to use SSD in traditional arrays. However since then they have been playing catchup with the rest of the storage marketplace. The XtremIO offering won’t set the world on fire; many startups are way ahead of EMC with faster, more efficient and mature products. However, inevitably for many existing EMC customers deploying XtremIO will be seen as an easier option than taking a risk evaluating the wider market. When the product announcements do come, you can be sure the EMC Marketing Machine will be in full swing.
Comments are always welcome; please indicate if you work for a vendor as it’s only fair. If you have any related links of interest, please feel free to add them as a comment for consideration.
[Note: Point 4 in The Standard Counterargument has since been added after private feedback]
- Netapp: The Inflexibility of Flexvols (12,306)
- Windows Server 2012 (Windows Server “8″) – Storage Spaces (11,468)
- Enterprise Computing: Why Thin Provisioning Is Not The Holy Grail for Utilisation (9,669)
- Comparing iSCSI Targets – Microsoft, StarWind, iSCSI Cake and Kernsafe – Part I (7,506)
- Windows Server 2012 (Windows Server “8″) – Virtual Fibre Channel (7,375)
- Review: Compellent Storage Center – Part II (7,333)
- Why Does Microsoft Hyper-V Not Support NFS? (6,724)
- How To: Enable iSNS Server in Windows 2008 (6,277)
- Data ONTAP 8.0 – Part III (5,750)
- How Many IOPS? (4,710)
- Comparing and Contrasting All Flash Arrays – All Vendors (18)
- The Virtual Machine is a Legacy Construct (7)
- Comparing iSCSI Targets – Microsoft, StarWind, iSCSI Cake and Kernsafe – Part I (5)
- Will Connected Data be Good for Drobo? (5)
- Enterprise Computing: Sun/Oracle Kicks Hitachi To The Kerb (4)
- Nasuni Cloud Mirroring – Protecting Against Service Provider Failures (3)
- EMC Megalaunch – Speeding to Lead Balloon (3)
- Virtualisation – Solving the Storage Problem (3)
- Hitachi Attacks Migration Costs with Non-Disruptive Migration Feature (3)
- Violin Memory Delivers Converged Storage (3)