When you utter the words “storage virtualisation” you can expect one of three reactions: 1. nothing — a vacant stare representing unfathomable depths of non-comprehension; 2. A visible bracing of the body and a carefully expelled sigh; 3. A slackening, like the person has just been kicked behind the knees accompanied by eye rolling and a plea to heaven for strength.
So maligned is the concept referred to as virtualisation, and so considerable is the money invested it by vendors, that IBM refused to partake in this article unless I agreed to a face-to-face meeting. “I hate the word virtualisation,” senior technical architect of Veritas Software,” Simon Elisha, said. “It’s a buzz word and means nothing. To explain it simply, nothing can beat a whiteboard. If you draw it up people usually get it in five minutes.”
A large part of what makes virtualisation so complex is that, depending on which vendor you talk to, the definition varies dramatically. From the perspective of Veritas and Computer Associates, it is a software solution. However, Hitachi Data System (HDS), IBM, EMC or Hewlett-Packard define it as a mix of both hardware and software.
At a business level virtualisation is about increasing the utilisation of storage capacity within an organisation. It is about consolidating the amount of servers required to perform functions, increasing redundancy of mission-critical applications such as key databases, and automating the management of storage assets to alleviate the workload on the IT department — plenty of good reasons for CIOs to be interested in it.
At a technical level this is all made possible by an additional layer of marts in the software and/or hardware that recognise the individual storage devices, not as standalone boxes, but as one wholesome data repository. In theory, a storage system with virtual capabilities doesn’t see the disk arrays or tape drives as individual boxes driven by their own operating systems. It is not restricted by physical demarcations. All it sees is the total capacity, or what is referred to as “logical volume”.
“Logical volume is a magical word,” Ian Selway of HP said. “It’s a volume that doesn’t really exist except as a conglomeration of a number of different drives.”
In addition, virtual smarts abstract the data. Rather than seeing it as a block which must be kept together, a system with virtual capabilities sees the data as ones and zeros constructed in a particular formation. Therefore, when it stores data it is not bound by the desire to keep it all together but can distribute it across several different storage arrays. For example, out of a one Terabyte file 30 gigabyte (GB) may be stored on an HDS Lightening array, 50GB on an EMC Clarion array and 20GB on an HP EVA.
The virtual smarts hold a table of the data so that it knows where it is and its reconstruction sequence.
“Most administrators like to know where their storage is being written,” Selway said. “It’s a kind of false security when they see the green light flickering on the array that the data is being written correctly and they can point to the box and says it’s there.
“What virtualisation says is that it doesn’t matter whether the data is striped across five or 52 arrays. Or if it is spread across this partial disk in that partial array. It’s irrelevant, providing that when you call up a file it comes back.”
This is a dramatically different way of thinking about data storage. A rudimentary form of virtualisation in servers has been around since the early days of mid-range computing. However, this new, more advanced aspect of virtualisation gives IT administrators capabilities previously only dreamt about in terms of utilisation and flexibility, which explains the hype cycle surrounding it for the last two years.
Traditionally, storage works on what is referred to as “physical volumes”. You had two terabytes (TB) of data laid out in a particular way, therefore you needed a specified amount of disk capacity all in the one array to accommodate it. This gave rise to a pattern of collecting server farms. For every additional function that a company required, for example if a new PeopleSoft database was installed, a new server was purchased with enough capacity to accommodate it. It proved to be an inefficient methodology with large amounts of capacity going unused.
“With virtualisation you can segment a RAID group into a number of logical units of data, or LUNs,” senior solutions architect for AlphaWest, Peter Marklew, said. “This is called slicing. It enables you to get better density of storage on the physical RAID so you can get down to using one-sixteenths of the arrays capacity.”
As well as greater use of the storage resources, virtualisation has the ability to hide from the user the complexity of what it’s doing. Rather than the IT administrator spending 95 per cent of his time tweaking capacity, the management layer decides how and where the data is going to be stored.
“The virtualisation layer controls where the data goes and holds the tables of where that data sits,” Selway said.
Marklew said that interoperability played a significant role. The ability of a virtualisation solution depended very much on how it was supported by the various OS (operating system) vendors, such as Microsoft’s Windows or Solaris or Linux.
“In a system what typically happens is that you have a pile of disk,” he said. “You set them up by nominating a bunch of physical drives and presenting them to the host (the server) as a RAID group. The host sees it and recognises it as a drive.
“The RAID group may also be segmented into a number of LUNs (logical units of data), for example, you might commit 40GB of the LUN to a Solaris system, 20GB to a Linux box and 30GB to a Windows server. For each allotment that you make you have to reconfigure the server and the array with corresponding instructions.”
It is a very time consuming activity and because of the direct connection between the array and the server, each time a drive fails or the administrator needs to make an adjustment to the system it has to be taken off-line.
“Most storage today is still direct attached storage (DAS) which means that if you want to add capacity you have to bring it off-line, add more capacity, reconfigure it and bring it back up again,” Selway said. “If it’s a simple operation you’re looking at two to three hours, but if you need to set up different RAID groups or slice the storage in layers it takes longer.”
One way to change that — and the reason why network attached storage (NAS) has grown so rapidly — is by taking the storage off of the servers and adding dedicated storage devices which you then point the servers towards. The servers do the writing and the arrays do the storing.
This is essentially what virtualisation offers — a modular technology which converts into a reduction in downtime and increased scalability. The servers still run the mission critical applications but because they are now connected to a monolithic data repository via multiple paths they can rely on the smarts in system to decide where the data will be held. When a disk or array is taken offline the server will send data via one of the alternate paths and when a new disk is added only minimal configuration is required for the server to recognise it.
‘Doing at a hardware level’
Depending on the vendor, there are different approaches to how this is managed, for example, in the case of an HP EVA the process is performed at the array via the smarts in the controller and away from the host device (the server).
Veritas on the other hand controls the process at a software level but then the server is making the decision about where and how the data is going to be stored.
“If you’re going into a heterogeneous environment containing older devices that don’t support virtualisation you can use Veritas software to add that functionality in for you,” Marklew said. “Given the choice though I’d do it at a hardware level if I could.”
What Marklew means by “doing it at a hardware level” is swapping out as much of the old machinery with non-virtual capabilities as possible and replacing it with new boxes containing virtual smarts. It is an approach favored by HDS, a vendor which believes that adding a new management layer of software without addressing the shortcomings of the hardware arrays is like sweeping the core issue under the carpet.
“The underlying problem is that the storage needs to be simplified,” said Vic Madarevic, storage solutions marketing and management for HDS, Asia Pacific International Americas. “By adding the virtualisation layer on top you’ll prolong the life of the system but at the end of the day the old boxes will have to go.”
“What we’re suggesting is the addition of a new hardware platform to consolidate the existing infrastructure. By putting all their storage requirements on to one of our platforms the customer can migrate many systems into one much smaller footprint.”
The suggestion sounds dramatic. Companies hate to spend money and telling them to dispose of one million dollars worth of infrastructure for the sake of standardising on one platform is enough to give them coronary failure. However, Madarevic said the approach carried a compelling total cost of ownership argument.
“Take into account the incompatibility issues of the different operating systems on both the arrays and the servers and the time it takes to manage that,” he said. “Then there is the consolidation aspect — the cabling complexities of the server farms, the floor space they consume and having to mask the inefficiencies of machines with different ages and pedigrees.
“Throw in the simpler management that virtualisation offers, through a graphical user interface (GUI), a more efficient use of bandwidth and higher levels of disaster recovery by being able to swap out failed drives and add capacity without taking the system off-line. Then add in the pure hardware expenditure of maintaining the legacy system — the SAN fabric, host BUS adapters, the number of switches and how many tape backups you need.
“Once you put all that into the equation you’re looking at a payoff inside one to three years. However, if we can’t prove that payoff, we won’t even suggest the mass consolidation approach to the customer. We’ll go for a gradual migration performed on a needs basis.”
Madarevic said a gradual approach should also be taken when switching on the virtual capabilities within the infrastructure.
“The customer doesn’t want to pay for where they will be in two years from now, or the amount of storage capacity they will require,” he said. “But by the same token in two years time they don’t want to pay a penalty for where they were two years ago. We’ll implement as much virtualisation as is appropriate at the moment, we won’t implement all the virtualisation capabilities that is in our storage.”
The benefit of virtualisation is the flexibility to buy 500GB of storage today and scale to 1TB in two years time without disrupting the businesses continuity.
Convincing the customer
While the concept of storage virtualisation may be curly to explain the benefits are ringing true with customers.
“Storage Area Networks (SANs) have gotten down to reality pricing,” Marklew said. “EMC has gone out there and stonked the price down so these days you can pick up an HDS 9530 with 500GB capacity for about $30,000-40,000.”
Selway said there was no doubt that the current climate was a buyers market with competitive pressure driving down list prices dramatically.
“There are some great deals to be had at the moment,” he said. “It’s a very competitive space because everyone’s chasing the same market.
Controllers have a list price of $300,000 for a one terabyte solution but no one pays that, the realistic street price is anywhere from $180,000 upwards.”
Meanwhile, vendors are looking to resellers to provide the voice of reason.
“You need a trusted advisor and a lot of businesses look to their resellers for that but it’s up to the resellers to get proper training,” Elisha said. Some integrators, such as AlphaWest have taken this role on with vigour.
“From AlphaWest’s point of view storage virtualisation about looking at the specific customer requirements and selecting the solution accordingly,” Marklew said. “Vendor’s like to say ‘Our solution does it all’ but we might use an HDS box on the one hand and a Veritas suite on the other.”
He said this often caused friction when an integrator supported multiple vendors because the excluded party would apply pressure saying ‘Why was our solution not presented?’ It requires the strength to push back and say ‘In this case its didn’t suit the client.’
“Once a customer has committed themselves to a vendor it’s very difficult to unentwine themselves because this technology sits right at the heart of their infrastructure. You have to make sure that you get the business case right, in reality, for the next 10 years.” Madarevic said integrators also needed to understand the hype versus the reality and be on the lookout for vaporware.
“Resellers have a responsibility to look at the technology roadmap and what the vendors are doing to support the customer,” he said. “They don’t want to dead-end their customer into a solution from which they will have to back track. A vendor should be committed to giving the customer choice and collaborating with the rest of the market on standards issues as opposed to a lock-in strategy for securing its ongoing revenue.”
To assist with the education process, HP has established three or four solution centres in capital cities so it can demonstrate the concept of virtualisation to customers.
Veritas also has a executive briefing centre and AlphaWest reported that 60-plus CIOs attended its recent Megabyte breakfast briefing on the subject of storage virtualisation.
This indicates that customers had not been completely scarred off by the gross levels of hype over the last two years and were now looking for its real value.
“Sometimes it can be as easy as saying ‘Here’s a Windows box, watch me add one gigabyte of storage to it’,” Elisha said. “Then the light turns on and they get it.”