SAN and NAS convergence: can satisfy storage requirements while reducing complexity and overhead. (Storage Networking).

Customers are deploying a variety of shared storage solutions in an attempt to reduce administrative costs, centralize resources and ensure high availability of business-critical data. Previously, the steady accumulation of servers and direct-attached storage (DAS) resulted in high administrative overhead and vulnerability to outages if individual servers failed. Shared-storage technology allows customers to consolidate both servers and storage, implement server clustering schemes for data access and dramatically reduce the per-gigabyte management costs by enabling fewer administrators to manage more storage. The two shared-storage technologies now available in the market--storage area network (SAN) and network-attached storage (NAS)--take different approaches in providing storage access over a network. Increasingly, however, customers who have deployed both SAN and NAS solutions are seeking ways to more tightly integrate all of their storage resources. The convergence of SAN and NAS over a common transport e nables customers to further streamline their storage operations while meeting the requirements of users for both SAN and NAS-based data.

SAN and NAS

The primary distinction between NAS and SAN is that NAS devices provide file access services to client computer systems, while SANs provide block data access. A computer may access files on a NAS device using file system device drivers such as network file system (NFS) or common Internet File System (CIFS) over TCP/IP. By contrast, a computer would access a SAN storage device using a SCSI block protocol such as Fibre Channel Protocol (FCP) or Internet Fibre Channel Protocol (iSCSI).

As shown in Figure 1, a traditional DAS configuration includes servers with parallel SCSI-attached storage devices. Due to this parallel cabling scheme, each server is the exclusive owner of its own storage, which may result in loss of data access if the server fails. In addition, since each server is capable of supporting a limited amount of SCSI storage, adding more storage capacity may mean adding more servers. Over time, this increases administrative costs and more administrators must be hired to manage more servers and storage units.

The clients access the file or application servers over the local area network (LAN), typically in a homogeneous operating system (OS) environment. Windows clients, for example, only have access to NT servers; Solaris clients only have access to Sun servers. Each operating system has its own unique file system format. NAS overcomes this OS-specific limitation by allowing cross-platform access to file systems. NFS and CIFS drivers on each client enable disparate OSs to access a single NAS device, thus enabling storage capacity sharing for heterogeneous computer platforms. For many customers, this is the main benefit of a NAS solution.

Because files ultimately reside on physical disks in the form of data blocks, a NAS device necessarily includes a block access storage component on the back end. With some NAS products, that block access mechanism may be based on parallel SCSI or ATA disks, while for others the back-end storage may be SAN-attached.

In many respects, a NAS device strongly resembles a traditional file server based on DAS. A client requests a file from the server, which in turn retrieves and assembles the data blocks that compose that file from disk. What elevates a NAS device above a traditional file server is that its server component has been optimized for efficient file access and stripped of auxiliary utilities common to most operating systems. A NAS device may be considered a more easily managed thin server running a streamlined OS, which in marketing parlance is referred to as an "appliance."

Describing a NAS product as a storage device requires a Clintonesque interpretation of the word "storage." A true storage device, such as a disk array, is the repository of data blocks. While a storage array may have onboard logic for aggregating data blocks (e.g. RAID), and although a file system's meta-data may also be written in blocks to the array, the transactional currency of the storage array is blocks, not files. Some intelligent entity (a file server, application server or workstation) must assemble those SCSI blocks into application-usable files or records. That an optimized NAS appliance fulfills this block-to-file assembly efficiently does not transform it into a storage device. A NAS device may contain storage, just as a standard file server may house its own storage disks, but for the purposes of political correctness a NAS device is in fact a file server, not a storage device.

A SAN is composed of servers or workstations and storage devices such as disk arrays or tape subsystems that are connected by a network infrastructure. The majority of SANs today use Fibre Channel as the storage network transport, although IP and Gigabit Ethernet SAN solutions are increasingly being deployed. The servers and storage devices on the storage network send and receive block data. For Fibre Channel SANs, the storage network is isolated from the LAN and is maintained as a dedicated network. For IP SANs, the storage network could be similarly dedicated to block storage traffic, or integrated with other IP traffic. In the latter case, virtual LAN (VLAN) functionality can be used to segregate storage from messaging traffic on a common IP infrastructure.

SANs enable sharing of block storage resources such as disk and tape. This overcomes the limitations of direct attached storage in terms of high-availability server clustering and storage consolidation, but is still OS-specific. Although a SAN may simultaneously support NT and Unix servers, each operating system will connect to its own storage resources on the SAN. SANs, then, do not provide the heterogeneous access characteristic of NAS, while NAS only provides this cross-platform access at the file level.

In terms of performance, a comparison of NAS to SAN is an apples-to-oranges exercise. A SAN may have multi-gigabit performance from disk to server, but that performance is measured in megabytes per second of data blocks, not files. After delivery of the data blocks, someone still has to do block-to-file assembly (or 'disassembly). A NAS device, by contrast, must retrieve data blocks from its own storage (which may be done over a high performance SAN), plus assume the processing latency of file assemblage. In addition, the NAS device incurs overhead in NFS or CIFS session management. If a customer needs raw speed, as in postproduction video editing applications, retrieving data blocks directly from disk will always be much faster than making a file request to a NAS device (which must then go get the blocks that compose the file). On a SAN, however, the real work is not done until the data blocks have been turned into something useful to the operating system, e.g., a video stream.

SAN and NAS Convergence

A single enterprise may acquire both NAS and SAN solutions to address different end-user requirements. An engineering department, for example, may implement NAS to solve cross platform issues. A human resources department may implement a SAN to support terabytes of employee information. Over time, integrating NAS and SAN connectivity may be desirable, if not for access to stored data then for consolidating storage backup operations.

A rudimentary form of SAN/NAS integration is shown in Figure 2. In this example, the thin server of the NAS device (NAS head) which is optimized for file delivery is attached to a SAN. Instead of housing its own storage arrays directly, the NAS device enters a shared storage network that facilitates addition of storage capacity as well as direct diskto-tape backup using third party copy (extended copy) capability.

At the same time, a SAN attached server could be responding to data base queries from clients and quickly retrieving the relevant data blocks that compose records from SAN-attached storage. Some storage vendors are integrating both NAS and SAN capability on their platforms by providing IP-based NFS and CIFS access as well as Fibre Channel connectivity in a single enclosure.

The configuration shown in Figure 2 contributes to reduction in administrative overhead and enables more efficient backup operations for both NAS and SAN, but still requires support of two separate network infrastructures. It achieves integration of SAN and NAS, but not convergence.

The introduction of IP-based SAN products enables further simplifying of this diagram. A NAS device uses mainstream IP networking for file transport. New IP storage protocols such as iSCSI and iFCP also leverage mainstream IP networks, but enable transport of data blocks. Convergence of SAN and NAS is thus facilitated by a common transport that can carry user requests for both files and blocks.

As shown in Figure 3, with both file and block storage data in P format, clients and servers can access the data type that is most appropriate for the upper layer applications being run. Since some iSCSI adapter cards support concurrent file and block access over a single Gigabit Ethernet interface, it is now possible to run both filebased and block-based applications concurrently.

The NAS device in this example may have its own direct-attached SCSI storage, ATA storage or use iSCSI to retrieve data blocks over the IP network. Likewise, an iSCSIattached server may store and retrieve block data from iSCSIattached storage arrays or traditional Fibre Channel arrays front-ended by IP storage switches. In addition, client workstations now have the option of making files requests to the NAS device, or going directly to iSCSI-attached storage for block data.

To make concurrent NAS and SAN support viable, interface cards should incorporate TCP off-load engine (TOE) technology to reduce processing overhead on the host systems. Currently, TOE-enabled Gigabit Ethernet cards can reduce host CPU utilization to less than 10% for iSCSI traffic and accelerate NFS and CIFS transactions.

Although this diagram shows a single IP network, the customer can chose to dedicate Gigabit Ethernet switches for storage switches or assign ports or blades of existing Ethernet switches, using VLANs to separate block data from messaging traffic if necessary. In addition, the IP network may extend, via IP routers, to any part of the enterprise to enable campus or remote access to either file or block-based data.

For the customer, the obvious benefit of SAN and NAS convergence over IP networks is the greater flexibility provided to support a variety of file and block-based business applications and the simplification and reduced administration required for the storage network. Implementing this unified solution, however, requires due diligence in network design. Introducing higher volumes of IP traffic for storage transactions must be accommodated by proper sizing of the network infrastructure to prevent excessive over subscription of bandwidth and unpredictable performance. The convergence of SAN and NAS, along with file-based and block-based storage virtualization, provides new opportunities for customers to satisfy enterprise-wide storage requirements while reducing complexity and operational overhead.

Tom Clark is director of technical marketing atNishan Systems (San Jose, Calif.) and author of Designing Storage Area Networks, P SANs and Designing Storage Area Networks Second Edition (published by Addison-Wesley)

www.nishansystems.com