iSCSI, the cloud, and the accelerator start to converge

As iSCSI gains traction in the market, it’s bringing about an interesting shift: the idea of storage networking, even if it’s not described in such terms, is becoming attractive to smaller businesses.

The growth of a new technology in that market inevitably creates challenges for the environment around it. With that in mind, SearchNetworking talked to HP and Riverbed Networks about the impact of iSCSI on the SME network.

Mark Nilsen, business manager at HP StorageWorks, says iSCSI is now a significant component of the company’s storage market. “Earlier this year, we saw iSCSI outperform, in terms of unit numbers, FiberChannel products, so there’s good acceptance in the market.”

With smaller customers now getting into IP-based SANs, Nilsen said, customers are starting to show an interest in features formerly considered as “enterprise-class” – “snapshotting, replication, clustering built into their storage solutions.

“We’ve had our HP StorageWorks Pout there 4000 out there for about two and a half years now. It allows things like split-site clusters – and we’re seeing even smaller customers doing this to replicate data across sites.”

That, however, is bringing new challenges. As iSCSI brings storage traffic onto business IP networks, it tends to break down one of the barriers separate SANs were designed to create: the division between traffic within the SAN, and “general” network traffic.

And that’s especially true in the WAN, where bandwidth is scarce: a company with a gigabit LAN might quite happily be an iSCSI user without noticing serious impacts. One with an ADSL2+ or SHDSL-based WAN, however, will certainly notice what’s happening once they split their storage across a couple of sites.

As Nilsen points out, redundancy within data lies at the heart of WAN acceleration. Documents, databases, Web pages, e-mails – all of these have a substantial amount of data replicated between supposedly different documents, and this replication is exploited by the WAN accelerator.

Damien Murphy, senior sales engineer at Riverbed Technology, describes storage as a “natural adjunct” to WAN acceleration, but it’s seen as a difficult problem: “can you deliver the same type of acceleration for purely block-based storage as we can get for applications?”

This goes beyond merely the use of iSCSI, Murphy said, which can fit well with an acceleration regime. When people think of what storage systems pass across a network, they tend to focus on the ability to read data and write data, and forget the third activity – that of booting an operating system.

And it’s that third capability, Murphy said, that’s going to pose the toughest problem for the intersection between storage and the WAN accelerator. Not only does the connection have to be fast: low latency is absolutely vital. “You’re not going to start booting machines remotely if you have tens of milliseconds delay between locations.”

Because it was designed to work across a very local cable, and didn’t have to share its capacity with LAN traffic, the SCSI protocol is also very chatty, making the idea of remotely mounting an iSCSI drive something that lives in the future.

Before we discuss the technology further, it’s important to ask why a remote boot capability such as this might be important.

Where customers are adopting cloud models, Murphy said, they often need to resolve a conflict between their desire to move activities out of their own data centers, on the one hand; and regulatory requirements to maintain control over their data, on the other.

“What if, for example, you wanted to boot virtual machines in a cloud infrastructure, but because of a regulatory requirement, you were required to keep the virtual machine images in your own data centre?”

The other cloud-style application needing WAN storage acceleration is the exact inverse: a company taking advantage of storage-as-a-service models, but needing to access that block-based remote storage from the computers in their own data centers using that storage.

“Either way, you want to make the cloud as elastic as possible,” Murphy said.

Restoration – especially from cloud environments – is another challenge in the world of block-based storage that Murphy said needs to be addresses.


Storage traffic, WANs and clouds: from the SMB to the enterprise


While iSCSI isn’t pitched at the very large enterprise market, he said, “the problems you have to solve for that protocol are the same kinds of problems that reach right up to the very large enterprise.”

Murphy cited one last use-case as an example: although minerals exploration is typically perceived as the kind of activity you find in very large enterprises, such organizations typically have lots of small remote sites.

Those remote sites, he said, generate lots of new, raw data out at the very edge of the network: the results of their geological scans, new maps, new analyses, all of which don’t fit the typical “lots of replicated data” assumptions of WAN acceleration.

“Today, people have to keep servers at the remote sites, because those sites generate lots of new data that’s not replicated, and is difficult to accelerate.”

When that problem is solved, Murphy said – when block data gets the same kind of acceleration as more mundane data types – then the storage can be concentrated back into the central site. That simplifies the remote site, since the engineers and geologists don’t need to maintain servers; and it would give data generated at the edge of the network the same kind of protection as data that always lived in the data centre.

Another driver, cited by both Nilsen and Murphy, is the never-ending growth in storage – something that hits the SME just as hard as the large enterprise.

In the SME, the never-ending growth in storage needs puts pressure on available space, and frequently means storage systems live in less-than-ideal environments, without the environmental protections that a bank can build into its data centre; while for the large enterprise, unending growth in storage is demanding both for cooling and ultimately real estate.

Again, the need is to get storage out into the cloud – and if the end-to-end trip can be in the block-based form that storage systems are accustomed to using, it becomes far less visible and therefore intrusive to the end users.

If, as Murphy expects, solutions arrive to these problems in the next year, the result should be a new phase of growth in the cloud services market.

Read more on SAN, NAS, solid state, RAID