In Person: Storage Technologies
'Our goal is to simplify storage'

From an engineering company with technical expertise, Network Appliance (NetApp) has evolved to an $800 million enterprise storage provider. In its 10 years of existence it has survived the dot-com crash (much of its business once came from dot-coms), and competed head on with storage giants like EMC and HP. Dave Hitz, Executive VP Engineering and founder Network Appliance, tells Network Magazine how he survives in a highly competitive market. He also sets the record straight on a few technical issues concerning NAS and storage virtualization. by Brian Pereira

The NAS market seems to be lucrative because a majority of enterprises are more interested in low-cost storage solutions. As a result, new players like Dell are getting into NAS storage. What's your strategy for maintaining leadership for NAS devices?
Our strategy is based on two things. Firstly, it's the Business Relationship. Customers want a high assurance, technical competency kind of relationship with the vendor. Our new competitors have not been good at establishing that kind of enterprise relationship (for storage and servers). Secondly, you need to invest in technology. You need to have a single solution for NAS, SAN, Unix and NT. The new competitors do not invest as much in technology as we do.

Can you elaborate on your recent alliance with Hitachi? Are you also considering partnerships with IBM and EMC? If so, what will be the nature of those partnerships?
Here's the relationship that we've established with Hitachi. NAS systems need to have storage at the backend. So our NAS capability ties up to Hitachi's SAN capability. Our NAS devices will be a front-end to their SAN. Many people have already purchased Hitachi equipment. They tell us they'd like to access their SANs through our NAS devices, so that relationship makes a lot of sense.

A similar partnership could be possible with IBM. EMC is our number one competitor so it's very difficult to imagine a business relationship with them. The issues here are less technical and more business related.

How did you convince Oracle to be your strategic partner for storage solutions?
When we first approached Oracle they said they weren't interested in new technology. They wanted to know how we could help customers save money. So we did a study with a (consulting) company called Input. Input did a test of Oracle databases running on top of storage from Sun, EMC and Network Appliance. And they found that with NetApp, the TCO was 25 percent the cost of EMC. We went back and showed this to Oracle. Three years ago Oracle had almost zero NetApp storage, today it has over 400 terabytes of storage on our systems.

There are two reasons why Oracle decided to work closely with NetApp: One is to save money. The second is Grid Computing. There are companies that use thousands of Linux computers and these all access the same storage.

Today this is used for technical computing: software development, chip design, oil & gas (seismic processing), Hollywood special effects. But Oracle believes this will soon be used for business computing.

Oracle talks about its 9i Rack program and the challenge of running it on thousands of Linux computers. The challenge is for these thousands of computers to access the same storage. So we do joint technical work for enabling multiple Linux computers with a network backend of storage (either NAS or SAN).

You once offered only NAS solutions. What prompted you to move to SAN solutions, and other related areas like caching?
There are two reasons for this. Many of our customers told us they had requirements for both NAS and SAN. They wanted us to be their vendor for both types of storage.

The second reason is the technology. If you look at the technology we have developed for managing files (snapshots, data replication, remote location), it's similar to the technology for virtualized environments.

People in the industry are talking about features they wish they could have. We have already developed those features for NAS. If we could adapt those capabilities for SAN we will already have the features that people want.

Can NAS and SAN co-exist on the same network?
Today people do not run NAS and SAN on the same network. NAS is a TCP/IP based solution while SAN works on fiber channel. When people say they have installed "a big SAN," they really mean they have installed a fiber channel network.

So you need two separate networks for NAS and SAN. Five years from now people will not want to have separate networks. Ninety percent of applications can run on both NAS and SAN. Typically SAN is used for high-end applications. There are a few applications designed to run specifically on SANs.

Then there are other applications that are more suitable for NAS. For instance, home directories for a thousand users just has to run on NAS.

Because there are a small number of applications that run only on SAN or only on NAS, it presents a management challenge for people who want to run both these types of applications in their environment.

The determining factor for most people (in deciding on NAS or SAN) will not be technical. The decision will be influenced by business planning/requirement. Of course, it also depends on the existing infrastructure for which they made huge investments in the past.

A problem with NAS devices is that they take up too much bandwidth, especially during backup and disaster recovery. What's the solution?
There are two solutions. NAS systems (like ours) can attach to a fiber channel network for backup. The second solution is to have dedicated bandwidth for backup. Some customers will connect to a fiber channel network for backup while others will buy a new switch, and build a dedicated network over TCP/IP, for backup. You can attach a tape device directly to a NAS solution.

Many vendors and users are talking about Storage Virtualization these days. What's needed for virtualization to become widespread? What does NAS virtualization bring to the table?
The old thinking was that for high performance computing you needed the shortest and fastest path to the disk. The new thinking is that performance is not the most important thing because all the new computers are fast. The more important issue becomes manageability and flexibility as you change your business.

So everybody is starting to ask questions about Virtualization. They ask what layer of software is needed to run on top of all these disks to make them look like one virtual pool. If you look closely at virtualization, it has very much the same capabilities that file systems already have.

To fully understand storage virtualization one must understand the three things that could go wrong with disk drive storage. Firstly, it's not big enough. A customer's storage requirement exceeds existing capacity. The solution is to take a large number of disk drives and combine them into a single file system. The customer would try to manage the various disks as one large storage pool, and that's storage virtualization. With our solution you take a large number of NAS filers and put them on the network, then make them all visible as a single storage pool.
Secondly, the disk drive is too big. There may be a thousand users and each needs to save a few MS Word documents, which are very small compared to the size of the disk drive. So I would like to take a small number of disk drives on a system and virtually partition these so that I can allocate sufficient storage to each user as per his requirements. The solution is to take a single system and chop it into pieces, and say this part is for project-1, this is for project-2, etc—and hand out the data that way.

The third problem is heterogeneous storage. Some customers have solutions from different vendors and find it too complicated to manage. The way to manage all these different storage solutions is to follow open standards for NAS and SAN. We can solve this problem by following open standards and working with the partners (who are into data management). We have partnerships with IBM Tivoli, CA Unicenter, HP OpenView, and many smaller companies.

More open standards are needed for virtualization to take off.

Brian Pereira can be reached at brianp@networkmagazineindia.com