Virtual target
- Article 1 of 2
- Information Unbound, October 2006
Virtualisation is the key component necessary to deliver optimised storage area networks.
Virtualisation has been the watchword of 2006. With organisations looking to get the most out of their existing investments, the idea that servers or storage might only see utilisation rates of 15-20% has become an anathema to those in charge of IT budgets. IDC estimates that the amount of data storage being purchased by a typical enterprise is growing at a rate of 80% per year; using virtualisation to combine that hardware into a single resource can typically increase utilisation rates to 40% or even 60%.
While much of the virtualisation focus has been on servers, virtualisation of storage – and of SANs in particular – has attracted the interest of many organisations looking to get the most from their existing investment and to reduce future investments. Finding ways to reduce storage costs and increase utilisation rates has started to become very important. Although virtualisation is primarily of interest as a way of increasing storage utilisation rates, it can also be used as a way of implementing information lifecycle management (ILM) processes, dealing with peaks and troughs in demand, increasing security and improving business continuity.
“Virtualisation is a way to deal with the day to day problems of storage by getting rid of its ‘storageness’,” says Steve Legg, IBM’s storage virtualisation architect. “Organisations are looking for capacity and performance, not the behaviours associated with vendor x – its ‘storageness’.”
When a new project begins in any organisation, that project will need a certain amount of storage. Often, much of the storage admin’s task will be finding space on different arrays for the project – or recommending the organisation buy new storage. “Many people say to me, ‘Much of my job is putting little bits of storage together to make big storage’,” says Legg.
Storage virtualisation comes in many forms. There are routers and controllers from companies such as Cisco and Broccade that are capable of making a storage array or several fabrics appear as many different volumes, allowing organisations to isolate particular applications to particular virtual volumes. This can improve security, since different applications can no longer access the same volume, and reduce the chance of unexpected interactions between applications.
It is also possible to partition SAN fabrics, isolating problems only to ‘virtual fabrics’ when they occur and reducing operational load. However, James Opfer, research vice president, Gartner Dataquest Research, says the uptake of such devices is not that high so their use isn’t that common at the moment, despite having been on the market for several years.
Instead, virtualisation usually involves uniting rather than sub-dividing SANs and SAN fabrics. This can be done at many possible layers of the SAN, each of which has its own benefits and drawbacks. At the server level, it’s possible to install software that creates a virtual view of the SAN storage available. However, this isn’t as fast as a hardware-based system, is OS-dependent and only works for the individual server.
Virtualisation is also possible at the storage level by building it into the storage arrays. This form of virtualisation is faster than virtualisation at the server-level but not as capable at virtualising across SAN fabrics.
“Virtualisation should exist in the brain of the storage system, not anywhere else,” argues John Joseph, VP of marketing at EqualLogic. His company, which sells iSCSI-based SAN hardware predominantly to SMEs, has a patented virtualisation technology included in each array. When an additional array is connected to the fabric, the virtualisation system is able to spot the other arrays on the fabric and, depending on the settings, reorganise stored data, potentially migrating data from one array to another if it’s nearly full. If the new array has different capabilities, the arrays can negotiate to move the more frequently requested or more important data onto the faster or more powerful array, for instance.
However, virtualisation on the array tends to require proprietary systems, doesn’t work when arrays are on different fabrics, and in the majority of cases, doesn’t work with arrays from different manufacturers.
Fabric- or switch-level virtualisation is garnering the most interest since it’s possible to overcome most of these problems, even if it does require expensive hardware. “There’s general interest from my customers,” says Andy Holpin, storage solutions consultant at services company Morse. “I’ve been talking to one large customer who wanted a proof of concept: that it was possible to virtualise and migrate data between different types of storage, mirroring between an EMC and a Hitachi and so on.”
Virtualisation can either be done inband or out-of-band. Inband requires hardware that can sit between all the network’s storage arrays and servers and mediate data transfer between them using standard SCSI commands. If the hardware isn’t capable of moving data at the same speed as the rest of the SAN, this can affect scalability and throughput, so the requisite hardware has to contain considerably processing power, pushing up costs. Indeed, Morse’s Andy Holpin says that inband switch-based virtualisation raises concerns for many of his customers over performance and its ability to cope with future demands. Out-of-band, however, requires agents to be installed so is more rarely used, although EMC, for example, offers an out-of-band virtualisation appliance called InVista.
As well as pooling a set of arrays into a single resource, virtualisation can also provide a number of other capabilities, depending on the systems involved. “When you’ve 10 projects and want to allocate 1TB to each project, you can virtualise everything instead,” says NetApp’s consultant systems engineer, Dave Logan “You present your storage as 10TB when you only have 3-4TB at the back end, say, and then use a policy engine to expand volumes and delete snapshots accordingly.”
Virtualisation can also provide ILM capabilities using policy engines. By setting up rules for data importance, it’s possible for the virtualisation system to migrate data from tier 1 storage to tier 2 storage or archiving systems, while presenting the pooled storage as a single resource: the data will appear to be in the same location as before, but will actually be on a completely different and cheaper storage system. Alternatively, if an operator knows that certain data is accessed more frequently on certain days, it is possible to automatically move data between different tiers of storage to cope with the different levels of demand.
Of equal interest to many organisations is the possibility not just of getting more out of their existing storage but for seamless back-up to other systems using disk-to-disk replication. Instead of migrating data, the virtualisation system can copy data to other SANs or storage media while moving data on the primary SAN.
While all these technologies can bring additional business benefits to SANs, one of the bigger knock-on effects is in terms of savings – not in technology acquisition but management costs and staffing levels.
“There’s no question that for some customers, cost has been a big barrier to adoption,” says Dell’s enterprise marketing manager, Hugh Jenkins “We’ve been able to take cost down considerably over the last three years. But complexity has remained the big barrier.” By making storage easier to manage and automate, much of those costs can be reduced.
Storage virtualisation is by no means a panacea and implementing is far from simple. But when properly implemented, it provides a way for organisations not only to get more from what they’ve already got, but to make managing it simpler and more cost effective.