We had the opportunity to talk with Rich Clifton, NetApp's VP and GM networked storage business units. The topics ranged from the IBM relationship to tiered storage, maximising IOPS, storage intelligence in the network and de-duplication. What became apparent is that there are several strong lights underneath the NetApp bushels.

TechWorld: How is the IBM relationship working?
Rich Clifton: It's meeting our expectations in terms of what's happening in the overall market. It seems to be going along fine.

Comment: NetApp and IBM are learning how to work effectively together. This is taking time to accomplish and it's too early to say if the IBM channel is selling lots of NetApp product. IBM is not like a reseller needing a standard channel program set of carrots and sticks to perform. Nor is it a Dell with a single direct sales channel to market, a highly effective but one trick pony. IBM is far more than this and tuning the relationship on both sides is necessary before performance happens. It's a little like a SAP implementation in that lots of groundwork is necessary before the results flow through. That is the impression I derived from Rich Clifton's comments.

TechWorld: Can you have a single virtual pool of storage serving both Fibre Channel and iSCSI SANs (storage area networks)?
Rich Clifton: Yes. Consider that there is a physical pool of storage. Layered on that are the containers, the FlexVols. LUNs (logical unit numbers) and files are layered on that. A LUN can be accessed over either Fibre Channel or iSCSI. You can switch back and forth. You can have one mode of access as the failover for the other.

TechWorld: Can you have tiered storage inside ONTAP GX? (I know you can move files between primary storage devices inside the GX global namespace but can you move them from primary to secondary storage devices inside that namespace?)
Rich Clifton: In terms of where we are going, yes. In terms of ONTAP GX now, no. The GX product is focussed on a particular user paradigm, featuring a large scale technical application infrastructure. The idea of a secondary storage tier isn't part of that usage target.

We already do this transfer from primary to secondary storage with ONTAP 7G. It's a normal process for our customers there.

FlexVols (containers) can have different priorities - FlexShare - so that you can mix different data types on the same set of spindles, such as a performance-focussed Oracle database striped across many spindles and a set of capacity-focussed applications filling up the rest of the space. The different containers have different service levels from the same set of spindles.

Comment: NetApp can provide tiering inside an array with both Fibre Channel and serial ATA (SATA) drives inside the array. But NetApp can also provide one effect of tiering by having different areas of the storage in an array have different service levels although the array uses one type of disk. This resembles the Pillar Data Axiom idea of different SLAs (service level agreements) from the same drive set although it doesn't use Pillar's short-stroking technology.

TechWorld: Is NetApp considering the use of 2.5-inch disk drives?
Rich Clifton: Creating create greater and greater densities for power and space are targets for us. Increasing the number of I/Os from our storage is a target for us. I can't comment on future products.

Comment: No comment as in probably yes. Clifton said that it's pissible to get the same I/O performance from SATA drives as from Fibre Channel drives by having more SATA spindles than FC spindles. This idea of increasing spindles to increase the amount of I/O from an array of drives is understood by NetApp.

I feel confident NetApp has taken on board the Seagate message about increasing the number of I/Os from a shelf of drives whilst lowering the power consumption by using 2.5-inch drives rather than the current 3.5-inch form factor drives. Watch this space.

TechWorld: Should storage intelligence and management reside in the network? (The implied contrast here is with Brocade and Cisco who conceive of storage management intelligence running directly on fabric switches and directors or on an appliance connected to fabric switches or directors, as with IBM's SVC. Another example is DataCore's SANmelody which runs on a PC appliance connected to the Fibre Channel fabric or Ethernet LAN.)
Rich Clifton: We have a SAN management tool OEM'd from Symantec; ccStore - Command Central Store. Our long term strategy and vision is that the actual driver for sophisticated storage resource management is the application. Things like NetWeaver and service-oriented architecture (SOA) specify the storage service requirements they need in an application framework, and this is provided in an automated way by a storage management facility.

Comment: The idea is that the storage management application stack is split into two parts. On the server side of the network it is embodied in application frameworks. On the storage side of the network it is placed in storage controllers. The network as such has no storage management intelligence in it.

NetApp's storage management view is thus the third side of a triangle with the other two sides being fabric/network-centric - with Brocade, Cisco, EMC and IBM - and storage controller-centric - with Hitachi Data Systems.

TechWorld: What is NetApp's view of de-duplication? (Previously I have written that NetApp does not possess sub-file-level de-duplication. It turns out that this is quite wrong. It does as we shall see.)
Rich Clifton: There are three forms of de-duplication: (file) versioning which gives us the highest yield; stream-based compression as used with virtual tape libraries (VTLs); and block-level de-duplication.

We do have block-level de-duplication through an agreement with Symantec to use its NetBackup product. If it is a disk-targeted backup then we do block-level de-duping in the NetApp storage processor, in ONTAP. You'll see us field block-level de-dupe more over time.

The block-level de-dupe was in ONTAP 7G 7.1.0.

You will see other de-dupe technologies in our VTLs over time, but not block-level de-dupe.

Comment: Symantec's NetBackup doesn't do the de-duping here. It's NetApp technology doing it. NetApp doesn't provide de-duping for its virtual tape libraries. For one thing the NetApp VTLs don't run ONTAP. When the VTL is an intermediate store with the ultimate destination being tape then you can't send block-level de-duped data to tape. If the VTL is the ultimate storage target then de-duping its contents would make sense.

TechWorld: What's the NetApp view of archiving?
Rich Clifton: For NetApp an archive is a stage of storing data when its usage is long-term preservation with a low likelihood of access. Sometimes customers implement it on disk; some on tape and most customers still want this (tape). If the archive is on tape we'll facilitate that but we are not a tape provider.

Our VTL, through its vaulting software, is sometimes the end-target and sometimes not. If it is the target then you need to preserve the data in a spinning infrastructure. If SnapVault is the archive then you need to facilitate the compliance needs, i.e. WORM capability, deletion only after a defined retention period, and so forth.

The Kazeon IS products have an input here. With the help of indexing and classification you can decide what to keep for what period of time, i.e. FlexVols with specific properties for different retention regimes.

Compliance certification, by the way, is not for any product; it is for a customer's processes.

Techworld: Will we see Kazeon functionality added to ONTAP?
Rich Clifton As a general strategy doing more effective indexing in ONTAP is clearly a target for us. In terms of which partners and which technologies we'll use it isn't right to say. Having longer disk-storage data lifetimes and classification is clearly where we're going.

Comment: NetApp is coming up to its fifteen anniversary and is now a $2 billion-plus corporation. It has technologies that other vendors make much more marketing noise about: thin provisioning (think 3PAR); de-duplication (think EMC/Avamar and HDS/Diligent, and is carefully adding them to ONTAP, the O/S that runs across all of its products, except its VTLs, Decru encryption appliances and Kazeon Information Servers.

It has a quite distinctive take on the location of storage intelligence with, in crude terms, a dumb network connecting storage-enabled applications on servers with storage controllers at the back-end. Its strength is its single O/S environment across its core back-end storage devices.

A possible problem area is the provision of specific application versions of its server-side storage management. There isn't a single server-side storage management utility from NetApp. Naturally; there isn't a standard way for applications to talk to such a utility. The APIs needed aren't openly available. The production pace of such storage-enabled extensions to applications is limited and, obviously, the most important ones will be done first.