NASSAN Hybrids

NAS/SAN Hybrids

The purported section between file and block in networked storage may also be healed by the convergence of SAN and NAS architecture in hybrid storage products. As shown in Figure 5-8, a traditional NAS platform consists of a "head unit" that offers physical connectivity to both the IP network (the "front end") and to "trays" of disk drives (the "back end") configured into a JBOD or RAID array. The back end could scale within practical limits imposed by the number of accesses made to the storage across the IP network and the disk handling capabilities of the back end array bus and controller.

Figure 5-8. Traditional NAS design.

Given the traditional design of NAS products, capacity scaling was conceived as a "horizontal" process. That is, when storage capacity requirements exceeded the practical limitations of a single NAS appliance, users were advised to purchase another unit and install it on the network. This scaling approach worked reasonably well in small- to medium- sized environments where NAS was deployed for file storage " elbow room." However, as a scaling approach for files in larger shops , it created problems with respect to data management, as suggested by the NuView StorageX offering described above.

On the device management front, first-generation NAS suffered from the same problem confronting all storage: The more devices that were fielded, the more unwieldy the situation was from a management perspective. It wasn't until the early 2000s that leading vendors began fielding multiplatform management products to enable customers with hundreds of appliances to oversee and manage them from a single console. Smaller NAS vendors never provided a multibox management scheme, and instead used off-the-shelf web engines to enable their products to articulate web pages that provided access to configuration controls and status information. More than one storage administrator, confronting this proliferation of device-generated web pages, complained that the task of storage management had been reduced to the drudgery of "surfing the web" ”traveling from one device web page to another to obtain management information.

The dilemma of multisystem management was a partial driver for more robust and "vertically-scaling" NAS appliances. A more capacious NAS, in theory, could reduce the number of appliances that needed to be managed . . . or even fielded in the first place.

The other driver for vertical scaling (the ability to add more storage behind a NAS head) was the burgeoning data storage requirements of email, and, to a lesser extent, other databases that were being hosted on NAS. These applications were difficult to distribute across multiple NAS appliances as their data exceeded the capacity of a single unit. Adding in the multibox management issues, NAS became an increasingly problematic hosting platform for growing data repositories.

In 2001, work began on addressing the vertical scalability of NAS (and other direct-attached storage arrays). Vixel Corporation, a Fibre Channel fabric switch-maker that had found its position in the market marginalized by the more popular FC switch vendors of the day, led the charge to deliver chip-based FC switches that could be integrated into storage arrays to enable a fabric interconnect among disk drives inside the array. Vixel chips and others found their way into NAS in late 2001 and early 2002, affording the capability to attach a greater number of drives to NAS heads, and increasing NAS capacity vertically in the process. This architecture (see Figure 5-9), combined with improvements in the NAS OS, effectively merged NAS and SAN technology to create a highly scalable appliance well-suited to certain applications, including e-mail.

Figure 5-9. Vertical scaling in NAS facilitated through FC switching.

The approach, while affording an increase in capacity, did little to address the practical constraints imposed on NAS storage by the protocols used to read and write data stored to those devices. Experts estimated that upwards of 70,000 instructions needed to be processed in order to open a file or bit location on a NAS disk using the Network File System (NFS) protocol. As a result, the more concurrent accesses made to the NAS appliance, the slower it operated. Moreover, the more capacity you added to the NAS, the more likely you were to see increased concurrent access attempts. So, latency became a constraint on NAS scaling.

Various approaches are being tried as of this writing to rectify the situation. Some vendors seek to place most NAS OS functions into faster-executing silicon and to add large memory buffers and caches in order to facilitate improved request handling speeds and feeds. In the early 2000s, work was done on the creation of a zero copy protocol called the Direct Access File System (DAFS) that would enable memory-to-memory mapping between initiators of I/O requests and their storage targets. This protocol, plus the work being done on Remote Direct Memory Access (RDMA) over TCP/IP, promise to dramatically reduce latency by reducing the number of steps required to request, process, transfer, process, and use data from storage platforms.

While zero copy protocols continue to develop, so does the hybridization of NAS and SAN. To facilitate high-performance database hosting on NAS appliances, architectures are in the works to enable, not only network file system protocol connectivity, but also network block protocol connectivity, on the NAS head. Such a hybrid might take the form of the platform described in Figure 5-10.

Figure 5-10. NAS/SAN hybrids with Network File System and Network Block Protocol accessibility.

In the figure, the back end of the appliance is vertically scalable using FC fabric or iSCSI network switching. The head end has been enhanced to support traditional NFS-, CIFS- and HTTP-based file access, and also iSCSI access for block storage. This enhancement effectively reunifies block and file storage in a single storage platform, a trend that began with Microsoft's introduction of Exchange Server 2000 (which required a block channel), and that will likely be underscored by the increasing substitution of object-oriented databases for traditional file systems going forward.