Smart networks: embedded devices and intelligent storage - Storage Networking - Industry Overview

Computer Technology Review, Nov, 2002 by Christine Taylor Chudnow

Embedded devices add levels of intelligence to their hosting computer or smart device, and are a common technology on telecom and data networks. But they have been trickier to establish over storage area networks, even though there are clear advantages for using intelligent embedded devices at the hardware layer.

There are several ways to dissect a SAN. A common approach is to think of it as two layers: a physical layer with subsystems, NAS appliances and interconnect technologies, and an application layer that manages, protects, and provides access to storage resources and stored data. The hardware layer largely controls speed, performance and scalability, so adding virtualization and policy intelligence at this level would allow the SAN to operate storage pooling and application provisioning quickly and efficiently. But the reality is that embedding intelligence in SAN hardware components is an awkward process. SAN data paths traditionally follow a master-slave model between the host and its attached storage devices, which are most often RAID subsystems. The subsystems are expected to wait for instructions and incoming data from attached servers, and only swing into action when they receive the data. Hardware-based storage control is more reactive than proactive, as storage administrators use zoning and LUN masking to protect subsystems from warring servers. (Think of it as Server Wars: In environments where two or more servers use the same subsystem, Server A may try to grab every individual disk for itself, or Server A and Server B may blithely overwrite each other's data on the same disks.)

Role of Intelligent Switches Has Grown

Given the master-slave architecture, most storage vendors concentrate on managing storage at the software (host) level. But in spite of these inhibitors, there is a promising development around intelligent switches. Embedding intelligence in physical layer switches would move SANs from a shared storage architecture to more dynamic switch-based architectures. These switches operate now at the individual subsystem level, and in 2003 promise to operate over the entire storage area network.

The intelligent switch dates back to the 1980s, where it first appeared in voice networks. It became active in data networks during the 1990s, where it automated functions and eventually morphed into sophisticated routing switches. Within the storage network, vendors hope that intelligent switches will eventually help the SAN to evolve from its present shared model-where hosts share their attached targets on a limited basis-to a policy-driven, virtualized switched network. To accomplish this, intelligent switches are combining switching architecture with I/O routing intelligence.

Switches are already crucial to storage network fabrics, and serve as edge devices that provide connectivity across multiple networked devices. But as so-called Layer 2 devices, they cannot sense or control higher level functions on the application layers. However, just as data networks evolved from Layer 2 switching to Layer 3 and then to full-fledged routers, so storage network switching is adding intelligence to the storage mix. Intelligent switch development is producing both intelligent switches embedded in individual RAID subsystems, and in mega-switches that can intelligently manage arrays across an entire fabric, providing virtualization and policy-based services. (These switches are not meant to replace storage management software, but to enhance it. They will be able to apply software-based policies and virtualization across zones while efficiently utilizing existing bandwidth.) The Gartner Group expects SAN switching products to grow from approximately $1.2 billion in 2002 to $4.3 billion in 2006.

RAID subsystems share the SAN with JBODs and tape libraries. But RAID arrays comprise 90% of SAN storage devices and are at the focus of intelligent switch development. RAID arrays house their disks behind a controller card that manages server requests. But RAID arrays have limitations centered in I/O issues. When a server issues a command, unless it is served in the cache it travels Out of the server's HBA and through the network until it reaches the target subsystem. The request enters the controller which then connects the appropriate disk, retrieves and processes the data, then sends the completed data back to the requesting server. Since multiple requests may be hitting the array at the same time, the array-even with dual controllers and multiple channels-is still subject to a single internal bus. This can lead to data latency and serious congestion under heavy I/O conditions.

One approach to solving this problem is to embed intelligent switches in the arrays that take over the data routes into the RAID subsystems. Another approach is to place intelligent switches into the fabric to accomplish storage network-wide switching abilities and virtualization. Both subsystem manufacturers and switch vendors are active in this market. Pirus's PSX-1000, for example, works within the fabric instead of at the edge, allowing storage administrators to reconfigure storage network file servers on the fly. The switch works with traditional Fibre Channel switches and supports multiprotocol convergence, including storage based on SAN (block), NAS (file), and IP (iSCSI). For internal RAID switching, Hitachi Data Systems (HDS) uses its Hi-Star internal switching architecture to combine disk, cache and server interfaces. This results in high internal bandwidths and capacity on subsystems such as its high-end Lightning series.