everythingpossible - Fotolia

NetApp: Hybrid cloud storage is a big deal; flash is not

At its Berlin Insight event, NetApp says hybrid cloud is the big storage story, with flash set to disappear from prominence

This week at its Insight event in Berlin, NetApp is boosting its cloud-centric vision of the future. In this NetApp view of the world hybrid cloud is seen as the way most enterprises will relate to the cloud in the coming years.

It’s a world in which organisations have multiple options regarding where data resides; on-premises and off, with data movement and processing occurring across locations, and with potentially multiple external cloud relationships.

We also live in a storage universe that has been dominated by the rise of flash storage. NetApp has always had an air of coolness towards solid state, from the early days when it proclaimed the only place for flash was cache. Even now its flash options are limited to a few models.

But that could be set to change. With FlashRay and its Mars operating system (OS), NetApp hopes to push flash to use with the use of triple level cell (TLC) flash, which will offer much higher storage densities and lower-cost solid state storage.

At the same time, however, NetApp says flash just isn’t that much of a big deal and that the overarching storage story of the next few years will be the cloud.

We spoke to NetApp founder and executive vice-president Dave Hitz, and Val Bercovici, senior director at the office of the CTO of NetApp, who fleshed out the company's vision for hybrid cloud storage and the future of flash.

Read more about NetApp

Computer Weekly: What’s the central message regarding NetApp and the cloud?

Val Bercovici: The “data fabric strategy” is NetApp’s vision for a hybrid cloud reality. It is a data-centric approach to a world where enterprises need to manage data in a multi-cloud, multi-vendor environment and is built on three main components.

These are: A rich data services layer with snapshots, clones and application, operating system and database integration; a data transport layer centred on SnapMirror and SnapVault; and a data-management layer with global policies and the concept of service level objectives and service catalogues.

Computer Weekly: None of these elements on their own are new. Are you saying they form the building blocks of your cloud-oriented strategy?

Bercovici: They are parts of a cloud-oriented strategy that fits with a world in which a minority or small majority of enterprises is operating in the cloud.

The data fabric strategy is about taking control across public and private domains in a multi-cloud strategy that includes Microsoft Azure, Amazon and IBM SoftLayer. When these come to fruition in terms of enterprises being on board, a data-centric strategy will be the only way.

The improvements to Data ONTAP, the addition of Cloud ONTAP, NetApp Private Cloud, OnCommand Cloud Manager – all these components form our hybrid cloud strategy which covers management at the customer premises, a transport layer between endpoints and a data services layer. They’re spread about but there’s a cohesiveness to the overall strategy.

In test and development phases the cloud is well-suited. There you need the elasticity cloud can offer. But production data, with its predictable workloads, sees the economics of the cloud break down 

 Val Bercovici 

Computer Weekly: This is a strategy that recognises there are opportunities in the cloud for a storage supplier. But isn’t it also based on a recognition of the limits of the cloud? It’s saying that the enterprise is not ready for cloud storage, that large chunks of data must remain on-premises for now.

Bercovici: I think it represents a maturity of the cloud. We’ve seen a diversification of approaches in use of the cloud. As organisations have adapted to the realities of the cloud we’ve seen a realisation about which workloads and datasets are most suited to cloud operations.

There is no one-size-fits-all; certain apps and data work well in the cloud and others don’t.

In test and development phases the cloud is well-suited. There you need the elasticity cloud can offer. But production data, with its predictable workloads, sees the economics of the cloud break down. Elasticity is not needed and any IT department can build the infrastructure needed better than a cloud provider.

We’ve seen real-world examples where organisations have operated test and development in the cloud then moved to on-premises when workloads become predictable.

The data fabric strategy addresses this with a uniform approach across the data lifecycle and ease of data portability between public and private domains.

Computer Weekly: If the hybrid cloud is a recognition of the limits of the cloud, then what will happen when those limits on latency, concerns over security and so on are no longer there? That is, when enterprises can use the cloud directly?

Dave Hitz: We’re doing our best to prepare for the full cloud, with, for example, Data ONTAP in the cloud. But we don’t expect there to be a direct evolution to the full cloud; hybrid will be here for some time.

Look at what happened when Windows NT arrived. People said Windows would mark the end of Unix/Linux, but what we’ve seen is a migration back and forth between the two for certain workloads. So, what we’re expecting is something like that movement between hybrid and full cloud.

If you’re running a machine and keeping it fully occupied and will need to do so for the capital lifecycle, the customer can do that more cheaply out of the cloud.

The cloud is good when you don’t know how long you will run something and when you need to ramp capacity up and down.

But, to use an analogy, it’s crazy for me to own my own mail server when I need about 100th of one. And that’s what it’s like. SMEs especially can use the cloud, but large businesses do some of what they do best on their own infrastructure with some things out in the cloud.

So, hybrid cloud will be around for a long time yet.

The flash story will soon be over but with cloud we are just at the beginning 

 Dave Hitz 

Computer Weekly: Where has FlashRay got to? What form will it come in?

Bercovici: FlashRay is shipping in limited quantities and will add dual and multi controller iterations in 2015.

A key feature we’re keen to highlight is pending support in the Mars OS component of FlashRay for TLC flash and that’s due to the serendipitous fact that WAFL – NetApp’s file layout technology in its Data ONTAP OS – is based on a log structure file system, and that’s exactly what you need for efficient flash operations.

It’s a universal but well-understood characteristic of flash storage that the programme-erase (P/E) cycle has issues with write amplification. Almost all controller software for flash is based on a log structured file system (LSFS) as a way of dealing with this. And so is WAFL, which allows us to partner with flash suppliers and to write directly or natively to the media and do away with the overheads required for wear levelling.

Write amplification means that, for other suppliers, a 4k page write actually means multiple 4k page writes. They create unnecessary amounts of P/E cycle activity because they don’t have an LSFS.

And in TLC these problems are several orders of magnitude more problematic, but we can support them.

Computer Weekly: NetApp’s making a big deal of the cloud in a way it never did with flash. Why is that?

Hitz: The difference between flash and cloud is an illustrative one.

Flash only requires the customer to wait. Storage products will improve and become faster and better as flash is incorporated into them. The customer doesn’t need to think about it.

Cloud feels different. Customers have to make decisions; should I build my own infrastructure or use the cloud; which apps do I entrust to the cloud; will this provider provide what we need and will it stay in business? These are some of the vital decisions customers must make.

For those reasons the flash story will soon be over but with cloud we are just at the beginning.

Read more on Cloud storage