Most enterprises are taking a cautious approach to committing their precious data to a remote datacentre run by a third party. We examine the technology trends that, in time, could drive enterprise data towards the cloud
Many businesses are experimenting with cloud-based services that offer plenty of storage. But scratch the surface of the cloud and you find enterprise computing and enterprise storage. It may be architected differently, and it's a different expenditure model — but as far as Amazon, Google and Microsoft are concerned, they're renting enterprise facilities at a distance.
Despite the attractions of today's cloud services, most enterprises, faced with limited network bandwidth, no strong security model and ill-defined storage performance, are going to keep their mission-critical data in-house rather than in-cloud for the moment.
So what will it take to build a credible enterprise storage cloud, and what should enterprises be doing in the meantime?
Several storage-related issues have received a lot of coverage recently, but for various reasons should prove less troublesome than expected.
The threat of power-induced meltdown on enterprise storage strategies has been widely broadcast for years now. But before you recast your storage architecture on an ultra-low-power model, consider the fact that, according to research company IDC, power requirements are going to level off by 2013 or 2014.
The reason, says IDC, is that recent hard times persuaded everyone to take cost savings seriously — particularly in using storage more efficiently.
Meanwhile, 2.5in. hard disks and solid-state drives (SSDs) use a lot less power, and virtualisation continues to push power requirements down as ever more capable mainstream chips run more tasks for the same or less wattage.
Eventually, the analysts say, the pause in power consumption growth will end and the numbers will edge up again. By then — the end of the decade — faster and cheaper networking will make it easier to swiftly share data around the world, following the sun or wind, and a number of advances in materials science promise to kick in and cool things down.
There have been dire forecasts that hard disk storage densities have peaked. The last big breakthrough was the discovery of giant magnetoresistance in the late 1980s, which pushed hard disk drive storage densities into new areas. But we're close to the 1 terabit per square inch limit, which could close developments down by 2013.
The continued existence of magnetic tape seems to annoy those people who see it as a 60s throwback and unworthy of the modern world. Yet nothing can touch it for cost-effective, power-friendly, scalable and secure secondary data storage.
Nothing can touch magnetic tape for cost-effective, scalable and secure secondary data storage.
Tape has error levels orders of magnitude lower than the disks it backs up. It also has a new open standard called LTO (Linear Tape Open) to go up against Quantum's DLT and Sony's AIT.
Tape's role is changing. With deduplication managing to compress data by ratios of up to 20:1, backing up data to disc has become more popular, especially as restoration can be a lot swifter. However, tape continues to be particularly well suited to off-site archives, and is reaping the benefit of an open standard helping to drive down prices.
A combination disk and tape backup regime will work for a lot of people for some while yet. You'll need an hierarchical storage manager, but the maths remains true: tape is the best way to keep a lot of data very safe.
Current storage trends
If there's one overriding trend in enterprise storage, it's the fact that you're going to store more data next year than you did this year. And with IDC saying we'll be stashing 0.988 zettabytes next year, the numbers are, frankly, freakish (a zettabyte is a million million terabytes).
With this in mind, the trends at the heart of enterprise storage are surprising: here, the specialist solution is giving way to the standard, and the benefits of highly tuned, thoroughbred architectures pall in comparison with ideas that, not so long ago, seemed almost contemptibly inappropriate.
Ethernet vs Fibre Channel
Fibre Channel has more than 20 years of being the network designed purely for storage. It has supercomputer heritage, it had a head-start on gigabit speeds and it's thoroughly established in Storage Area Networks (SANs), where it moves things around swiftly and efficiently, allowing you to manage your storage with exquisite precision.
Ethernet had a messy childhood, coming up through personal computers and innumerable standards battles.
Ethernet, by contrast, had a messy childhood, coming up through personal computers and innumerable standards battles. It wasn't deterministic and it wasn't particularly robust, but it sacrificed being very good at one thing for being just about OK at many. Which made it good for Network Attached Storage (NAS), which looks like the sort of PC network that accepted all the inefficiencies of being strung together with high-level protocols.
Then two things happened. First, driven by its widespread use in the internet, Ethernet got better and easier to use, and the economics for using it to glue everything together became unanswerable. Second, Fibre Channel learned to work with Ethernet, with FCoE (Fibre Channel over Ethernet).
Now, 10Gbps Ethernet is more efficient at moving data than 8Gbps Fibre Channel. Because of differences in the way they code the data they carry, 10Gbps Ethernet can carry a real throughput of 9.7Gpbs, whereas 8Gbps Fibre Channel can only manage 6.8Gbps. So Ethernet can carry all of that 8Gbps Fibre Channel data plus three lots of 1Gbps non-storage traffic, on one physical link. And it can use the same infrastructure as the rest of your datacentre.
So the trend is in Ethernet's favour, as is the way that per-port costs are falling for the upstart technology. Look at what Brocade is doing with devices like its VDX 6720 Data Center Switches — melding virtual switching, 10Gbps Ethernet and FCoE into one simplified fabric that looks to the network like a single switch. That's a big statement for a company with a hefty market share in Fibre-Channel SANs.
Block vs file
Talking of old battles, how about SAN's blocks versus NAS's files? SAN is the incumbent in the datacentre, with NAS coming in from the outside world.
Traditionally, if you wanted a high-performance storage system, you built it as a SAN with its own dedicated network, you optimised it and you relied on the high levels of reliability and efficiency guaranteed by the vendors. All that's still true, but it requires a high level of expertise — often about a particular vendor's product line. It also becomes difficult to manage in a highly virtualised system.
NAS moves more intelligence to the storage system, making it easier to manage and share in a very dynamic environment where multiple virtual servers make complex demands. It's easier to scale up, and it handles storage entities as files and directories, which makes for easier automated backups, mirrors and so on. It's also a natural for IP networks.
Many established datacentres will be using a mix of NAS and SAN, and a generation of SAN experts will be around for a while. But the trend is towards NAS, as once again the workable, flexible technology uses market clout to break in.
Big or not so big?
So, does this move towards consolidation around concepts common outside the datacentre affect the future of vendors? It does. The same basic forces that are pushing commodity ideas into enterprise storage have already worked their magic in other markets.
In PC processors, it's down to one company and a second placer, after all sorts of niche markets that once supported technically superior designs sunk beneath the waves. Even the supercomputer world feels the pull towards that monoculture.
Already in enterprise storage, the rush towards consolidation, and the disinclination among many buyers to purchase anything that doesn't come from the biggest of big guns, has led to HP and Dell locking horns over who got to marry 3Par — with a huge premium attaching to its price tag as a result. HP walked up the aisle there, but Dell has recently named the day for a match with Compellent.
Meanwhile, Oracle knows that, as it stands, it's not up there with the leaders; however, its strategy in buying Sun and settling disputes with NetApp shows where it's minded to go.
Disturbing the equilibrium
Disruption in enterprise storage isn't just possible — it's essential, if we're to keep up with projected growth. Here are three disruptions that will drive the progress of enterprise storage.
The switch to solid-state
It's already cost effective for an increasing range of high-performance tasks to use a bank of SSDs instead of conventional rotating hard disks, even given the SSD's disadvantages of higher cost and shorter lifetime. But that's just the first phase: the disruptive effect of solid-state storage increases massively as we change the underlying storage architecture to make the best use of its potential.
Companies such as Fusion-io are making SSDs that look much more like fast memory tightly coupled to processors than ordinary disks. In effect, they're collapsing I/O down to the minimum number of layers running at the maximum speed. The marketing departments are pushing this as Tier 0. But it's early days, and sales have been held back because of worries over error rates and increased maintenance.
Yet flash technology improves by the year, as does that of the controllers that compensate for its drawbacks.
There are also some notable surprise runners: HP's memristor, for example, has the potential to leapfrog flash in performance and in reliability, as well as density and power utilisation.
Storage is no good if you can't get data in and out quickly enough, which is why most of the discussion around enterprise storage architecture concerns the networking fabric from which it's woven. For speed, you can't beat optical fibre. For cost, you can't beat copper.
Silicon photonics or silicon nano-optics combines the best bits of the two technologies. Demonstrated by Intel and announced by IBM, this takes all the expensive, cumbersome and inflexible aspects of optical fibre — the stuff that translates between light and electricity — and places them on-chip.
Lasers, switches, modulators, multiplexors, detectors — all of these components can be made of silicon. Once there, the magic of Moore's Law makes them cheap, simple and scalable. Both IBM and Intel have talked about terabits a second being achievable, and of course this would be very closely coupled with the other circuitry on the chip — memory, CPU, controllers.
Enterprise storage should be the first area to benefit from this, as it has the need and the margins to make economic use of the first generation of commercialised photonic products. Very cheap, very fast networking with the capability of traversing kilometres will be a game-changer — not just in terms of raw performance, but by enabling entirely new distributed topologies.
Search and security
How do you find data quickly among the petabytes, and shield it from unauthorised access? These two problems are related. At some point, the way in which filing systems and search algorithms interact will change fundamentally, as the storage system itself maintains more and more information about what it contain, and who should see it.
Ideas such as ZFS and Ceph are taking on the challenge of mapping a changing, scaling storage infrastructure to the needs of computation, while Google continues its ten-year experiment in creating the world's biggest, smartest database.
Security is lagging, but the model is evolving where each piece of data is bound with rules that describe how it can be distributed, and the infrastructure itself becoming permissive or impervious according to who requests what.
What the three disruptions will bring
Extremely fast local caching thanks to solid-state disks. Extremely fast long-distance networking from cheap on-chip photonics. And extremely smart data-centric fabrics that enforce strict security. With all this in place, we might just have an enterprise storage cloud that makes sense.