My last post discussed the issues with EMC’s strategy around the marketing of their new VNX platform. Some people will think picking holes in a vendor’s strategy is easy writing, so for those of that persuasion, I’ve taken the time to look at EMC’s portfolio and express an opinion on where EMC should go next.
As I’ve discussed, EMC has a wide range of storage array technology now.
- VMAX – branded as Symmetrix then DMX, this is the original storage platform that made the company big. Twenty years ago when storage was widely distributed in servers, Symmetrix brought a new way of working with consolidated storage and advanced features like RAID and cache. VMAX has done well over the years and still is a major part of EMC’s portfolio, but some of the technology features are looking a little old, with many layers of software being used to implement advanced features such as thin provisioning.
- VNX – this platform started life as the CLARiiON, when Data General was acquired by EMC in 1999. At the time, the CLARiiON was an advanced piece of technology, with dual controllers and RAID management functions. The platform was merged with the Celerra NAS gateway to produce VNX and up until VNX2 that heritage could still be seen in configuration dumps.
- Atmos – EMC’s attempt at scale-out low cost cloud storage. The platform had no frills (like RAID), which perhaps scared customers who were used to the traditional VMAX storage model. I’m not sure how successful Atmos has been but I hear very little about it.
- Isilon – Scale-out NAS. Acquired by EMC in 2010, this is a good example of future storage architectures using scale-out functionality
- XtremIO – Not officially a GA product (but almost there), XtremIO is EMC’s attempt at scale out all-flash. It is yet another acquisition, for a rumoured $430m.
- ScaleIO – an even more recent acquisition (July 2013), ScaleIO is a distributed scale-out multi-node solution that utiilises existing servers and storage resources.
- Project Nile – the latest EMC teaser – a block, NAS and object scale-out solution that can be purchased “on demand” and ships to the customer fully configured.
So what do we have here? Well, the legacy VMAX and VNX platforms are monolithic and dual architecture designs. These architectures don’t lend themselves well to the development of cloud and server scale-out architectures that we can expect to see in the future. In addition, they were designed when storage needed to be centralised for resiliency and performance. However, these days hard drives and server components are significantly more reliable than ever before and flash storage is delivering performance with high I/O density (IOPS per GB). That’s why scale out architectures such as SolidFire and Nutanix have been successful and why with Open Computing architectures we are moving to the server as the level of granularity.
In a recent post, Chad Sakac discusses the scale-out issue, but has to contradict himself in order to justify the existence of VNX2. Check out the paragraph and statement “scale-out is the future, scale-up is the past”. Unfortunately for Chad, as a marketer he has to toe the line on VNX because of the huge mass of existing customers who still believe in the VNX platform.
The clear message from EMC’s portfolio is that scale-out will be the future. The question is how to get there. Well, EMC have all the key pieces; Atmos provides the raw storage capability; XtremIO, Isilon and ScaleIO provide the architecture. In fact, if the XtremIO solution is so good, then surely it could be retro-fitted with hard drives to deliver a hybrid solution. Let’s think about that one for a moment. If you look at storage usage today, you will typically find that on any large shared storage array, 80% of the activity is being done by 20% of the data, with a long tail of small volume I/O. I talked about this three years ago in a post on Wide Striping. So, with that in mind, an array could be constructed using the XtremIO architecture with just a few SSDs and the majority capacity provided by large slower hard drives. The question is whether the architecture can target writes effectively. I’ve talked about that too, in this 2009 post on automated tiering. Using the XtremIO architecture as the basis of a hybrid array could work, or alternatively, taking the ScaleIO software and using other EMC hardware components could deliver a solution as well.
The Architect’s View
If EMC have all the pieces to evolve VNX, then why have they not done so? There are probably a number of reasons:
- XtremIO and ScaleIO are very early technology solutions and so not fully featured with things like snapshots, replication, clones, thin provisioning and so on. There would be significant development work required to get this into place and these are only recent acquisitions.
- VNX is a cash cow. EMC don’t want to kill off a platform that could keep revenues coming in for many years yet.
- Upsetting the ecosystem. Many organisations will have built processes and skills around VNX. Radical change to the platform would represent a major disruptive force.
The problem EMC have is that other vendors, have already made the move. NetApp (aka “Brand N”) have re-engineered Data ONTAP from the Spinnaker platform to take over from legacy ONTAP, the so-called 7-mode and cluster-mode versions. Although these were totally different operating systems, cluster-mode has been changed to support all of the previous 7-mode features. The only issue for customers will be migration. NetApp are hoping 7-mode will fade away gracefully as customers transition themselves over. HP have let the EVA die off in preference for the 3PAR platform, which now scales from entry level to enterprise and includes all-flash. It could well be that EMC have known these issues all along and VNX2 is merely a stop-gap to the release of a new platform that solves all these problems. Could this be Project Nile? We will have to wait and see.
- Speed2lead – Part1 – Next Generation VNX (virtualgeek blog)
- Enterprise Computing: The Benefits of Wide Striping – Avoiding a Long Tail
- Enterprise Computing: Automated Tiering – Why Move The Data?
Comments are always welcome; please read our Comments Policy. If you have any related links of interest, please feel free to add them as a comment for consideration.
Copyright (c) 2007-2020 – Post #C505 – Brookend Ltd, first published on https://www.architecting.it/blog, do not reproduce without permission.