‘Storage Hypervisors’ – IBM Proposes A New Industry Term

Storage Hypervisor Debate Highlights

• Improving resource utilisation is central to reducing data centre costs
• SAN and NAS have improved virtualisation, but typically within a single vendor’s range
• VMware has been highly successful in lowering server spending by heterogeneous applying virtualisation
• Virtualisation platforms, management software and application across multiple vendors’ kit required
• IBM’s own approach includes SVC and TPC – it can virtualise and manage most other vendors’ storage
• We may need a software-only vendor (as VMware is in servers) for storage hypervisors to take off

CIOs Place Strong Business Value In Server Virtualisation

In the late 1990s a handful of vendors proposed visions of Utility Computing where data centre resources could be pooled – reducing costs significantly by increasing utilisation. By that time Unix machines had joined mainframes in being fully virtualised, which proved very valuable in the large number of data centre consolidation and/or outsourcing projects introduced in the economic downturn of 2001-3.
Despite already being available on other CPUs, server virtualisation really took off when VMware addressed the x86 server market with its hypervisors from 2003 onwards. Along with Microsoft’s Hyper-V, Xen and (more recently) KVM, VMware’s server hypervisors have been successful in lowering spending on server hardware through increasing utilisation. Their affect can even be seen on the development of total market spending.
In the x86 world hypervisors allow VMs to be used across different vendors’ kit and developments in management software such as vCenter allows Virtual Machines to be moved from between different vendor’s kit as well.

Storage Virtualisation Differs By Vendor

The Utility Computing vision and even data centre consolidation movement also addressed storage, which suffered from poor utilisation by being implemented in multiple unconnected islands. However there has never been the equivalent of a VMware for the storage systems world. The increased utilisation of SAN and NAS is typically done within the confines of a single vendor’s range (and sometimes only a part). We can’t argue that the underlying components differ more than in the server world – disk drives from Seagate, Western Digital and Toshiba are just as standard as processors from Intel and AMD in our opinion. There are certainly parallels between VMware and NetApp – but the latter remains exclusively a software vendor, while the latter is not.
Clearly CIOs would welcome a parallel ‘storage hypervisor’ movement, if it could bring anything like the costs savings they’ve enjoyed in the x86 server space.

Virtualisation Platform And Management Needed For Parallel Movement

For an industry trend towards storage hypervisors we need a number of things. For instance:

• A virtualisation platform allowing arrays from of different types and from different vendors to be added together to a pool
• Management software to allow LUNs to be shifted across the pool easily

However there will be a difference in the nature of the hypervisor itself, which is a specific piece of code associated with or under a server operating system. In the storage world the ‘hypervisor’ itself will be more like an agent and certainly harder to define technically.

IBM’s Chances With SVC And Tivoli Storage Productivity Center

Of course IBM wouldn’t be proposing a new industry term if it didn’t think it was well positioned. It has the SAN Volume Controller (SVC) as the virtualisation platform and Tivoli (Storage) Productivity Center (TPC) as the management environment. It is one of the few vendors which have steadfastly insisted on addressing the inclusion of other vendors’ storage systems – insisting on Open Systems. These are not new developments, as in other areas it has ploughed its own course for many years.
On our recent call we also heard from Ricoh America, which enthusiastically endorses IBM’s approach, having introduced SVC in 2004 and TPC later. It has a total of 1.2TB of storage within this environment (85% of all storage), as well as three different types of storage (albeit all from IBM). It reports making significant cost savings through increased utilisation, moves LUNs around regularly and manages its operations without a single member of staff exclusively dedicated to storage. It also believes that IBM’s Open Systems approach minimises the potential of vendor lock-in. LV 1871 is another customer which speaks highly of he savings it has made through introducing SVC.

Stretching To Midrange And Beyond Arrays

IBM’s SVC remains a high-end solution and the concept of storage hypervisors will only be successful if (as in the x86 server market) it stretches to the midrange and eventually SMB areas. IBM’s Storwize V7000 is its first move, accommodating features of SVC and TPC. It believes the concept of storage hypervisors can stretch beyond blocks to tape and even from disk to spinning tape and that Amazon or Nirvanix type Cloud storage could be brought under common management.

Some Conclusions – Storage Hypervisors Are An Industry Issue

As a server analyst at heart I like the concept of storage hypervisor and understand why IBM wants to propose this as an industry term. There are strong parallels between servers and storage systems in the use of standard components, advances in virtualisation techniques and strong potential savings through improved utilisation. I have reservations in a couple of areas:

• Firstly – will users accept IBM as a leader in this movement? It is the largest system vendor and 100 years old – very unlike VMware which, despite its partial ownership by EMC, is viewed as a young agile player
• Secondly – storage systems get replaced faster than servers, due to the greater improvement in underlying component capacities; perhaps the more constant choices made by users favour the single vendor approach

Overall it would be good to see storage pooling fall into line with servers in making part of the Utility Computing vision of 10 years ago come true – but then we’ll need more from enterprise networks, peripherals and PCs.