Storage Virtualization: A Rose By Any Other Name
by Hu Yoshida on Mar 25, 2009
My last post on Storage Switzerland’s post on storage virtualization seemed to have generated some interest in the storage blogging community. StorageBod focused on my comment that Storage virtualization with the USP V can enhance commodity storage with the enterprise functions of the USP V controller. He comes to the conclusion that the USP V is “simply an array controller” and not truly virtualization. He offers the suggestion that we should sell the USP V as an array controller with the ability to attach any disk that provides the “best deal of the day,” and continues in his comments with ” .. if we look at them like that, they both (USP v and SVC) become less scary but also slightly less special but still extremely useful. ”
StorageBod (aka Martin G) makes a very good point about making storage virtualization less scary if we consider it as a storage controller that can attach and commoditize external storage. The only correction I would make is that the USP V is not “simply an array controller”. The term array controller came about with the introduction of RAID, Redundant Array of Independent Disks, when the primary function of a storage controller was to manage the RAID function and map them into LUNs. RAID is now done in chip sets on the back end of controllers and controllers concentrate on higher level functions like copies, moves, replication, load balancing, thin provisioning, and storage virtualization in the case of the USP V. With storage virtualization, the RAID processing is done in the external storage.
Barry Whyte of IBM also joined the conversation and agreed with both of us saying that “ they (USP V and SVC ) both do provide commoditisation at a fundamental level. SVC more so”. As Barry commented, there are very few times that Barry and I agree, but in this case I agree with him on both these points. Both the USP V and SVC provide commoditization at a fundamental level, and the SVC is more of a commodity since it is an appliance that lacks the resources and functionality of an enterprise control unit. He dismisses the diskless USP VM as a USP V without disks and ”still a fairly big box and quite expensive”. I tried to leave a comment on his blog, but I did not have the patience to register for an IBM ID , to post my comment so I am responding here in the event that Barry reads my blog
The USP VM is not just a USP V without disks. It is the same USP Architecture, but packaged and priced for the mid-range market in a 19 in rack. It has all the functionality of the USP V but scaled down from 128 processors to 32 processors. I understand that most SVC installs are packaged with 8 SVC appliances in a 19 in rack with two sets of FC switches for connectivity and the price is about the same as a diskless USP VM but with less functionality.
When Hitachi considered the introduction of the first generation of the USP in 2004, there was some consideration to calling this external attach rather than storage virtualization, but when we considered the definition of storage virtualization that was developed by SNIA in their 2001 white paper, we decided that we fit that definition. Specifically we could apply “virtualization to storage services or devices for the purposes of aggregating services or devices, hiding complexity, or adding new capabilities to lower level storage resources.” In other words we could commoditize lower level storage resources.
There was also discussion as to whether we wanted to do this since this would unbundle the storage media from the storage controller and allow customers to buy other vendor’s storage. Hitachi decided that this was going to happen as the natural disaggregation of storage whether we did it or not. In a way this would make us more competitive since we could compete in the storage controller market and the storage commodity markets with different cost structures, but with a common set of management tools.
So whether we call this storage virtualization or a storage controller with external storage the net effect is the same. It enables the enhancement of commodity storage with enterprise functionality without the need to buy expensive external storage systems.
Comments (3 )
I don’t equate ‘array controller’ with raid controller; I should have made that more clear I guess. In any modern array, the controller provides the brains and all the value, also the greater majority of the cost!
I think what HDS and IBM have done is very interesting and not without value; I think a more loose coupling of the array controller from the back-end disk is inevitable! Who knows, perhaps Cisco will simply build them into their UCS. I think that we live interesting times.
I read Storagebod’s blog and I can appreciate what he is saying – simplify the message. However, calling it storage virtualization is a good thing (I refer to it as external storage virtualization becase there are a number of internal storage virtualization technologies). For one, it allows HDS (and IBM) to frame a discussion. The conversation quickly turns into a very common sense dialog between HDS and the customer.
Based on Storagebod’s view then 3PAR and all of the subsequent vendors shouldn’t have called it thin provisioning and Data Domain should not have called it data deduplication. Having a specific term for a technology is critical to discussing and educating the market on its value.
Additionally, the HDS USP V is a storage system and has always been presented as such. It also supports external storage virtualization – which is a valuable and powerful capability that is unique. Customers often buy the HDS USP V in great part because of this capability.
I also disagree with Storagebod on his view of the technology – it is a big deal. And as we all know the invention of technology is probably just 1/3rd the battle – it is the testing, support, interoperability, Q&A and the field implementations that are also important. Especially when dealing with something like this – there are a lot of connecting points and they must all be considered and addressed.
The USP when it first came out had a head start because as Storagebod points out that it is a storage controller and it had many years of field proven support in that role (since the foundation of it was the Lightning). Today’s USP V now has years of experience in mission-critical environments performing storage virtualization and that adds to its viability and value to the customer. We often forget that technology and its applicability is more than just a feature but also includes all the required steps to make it efficient and functional in the data center.
First sorry for taking a few days to respond to your post and of course thanks for reading. I humbly disagree. While virtualization solutions can offer some benefit to commoditization, I don’t think it ranks very high in importance when compared to some of the other capabilities that virtualization can bring.
For a detailed response see: