Features - Enterprise Data Insights:

IS MANAGEMENT "IN YOUR FACE" ABOUT YOUR INTERFACE STRATEGY?
by Lee W. Payne, Overland Storage Inc

If you're currently researching interface strategies for your company, you're likely wondering which standards will win, which will die, or if they all will exist side by side. Perhaps you're feeling pressured by company executives to quickly develop a strategy that outlines the interfaces needed to move your organization forward and why.

Tape automation interfaces allow you to use a broad range of tape automation products that support both direct server attachments and shared storage attachments. Within today's modern enterprise data center, IT professionals use many different interfaces to meet specific applications requirements. For example, today's disk array often contains two different interfaces - a host side (Front End) interface to connect to a host or SAN / LAN, and a device interface (Back End) to connect to the actual storage device, in this case, the disks.

A tape automation interface strategy should outline plans for the next 24 months. Following is an overview of the different tape automation interface technologies that exist and their impact on the market.

ATA, Serial ATA Advanced Technology Attachment (ATA), a disk level device interface for desktop, workstations and laptops, is the evolutionary successor to the original Integrated Drive Electronics (IDE) hard disk interface popularized by the success of the IBM PC in the early 1980s. Because ATA costs 400 to 500 percent less than Fibre Channel or SCSI drives of equal capacity, over the next few years we will see increasing numbers of disk arrays incorporate ATA-based interfaces to deliver more cost effective storage to a price sensitive market. Serial interfaces simplify RAID controller architecture by reducing the board level routing nightmares associated with parallel busses. Serial busses also are easier and less expensive in switching environments. If it becomes available at parallel ATA pricing, Serial ATA will simplify RAID controller mid-plain architecture.

SCSI Small Computer System Interface (SCSI) has been adopted by all major OS vendors as the software protocol of choice for controlling block level access to data and has become the mainstay of enterprise class storage systems. Fibre Channel, for example, uses SCSI as an upper layer (FC-4 layer) protocol providing block level access to data. Because all of the block level protocols for Gigabit Ethernet including iSCSI, iFCP and FCIP use SCSI Command Descriptor Blocks (CDBs), SCSI CDBs will remain the key software protocol for enterprise class storage for the foreseeable future.

Parallel SCSI Parallel SCSI is the ideal interfaces for direct attach external storage with Low Voltage Differential (LVD). The maximum bus length of LVD SCSI is 12 meters. There has been a substantial decrease in demand for High Voltage Differential (HVD)-based SCSI products outside of legacy applications due to lack of support for HVD SCSI beyond the 40 MB/second rate. As a result, HVD SCSI has fallen by the wayside, surrendering most of the enterprise class attachment and clustering market to Fibre Channel.

Serial SCSI The next logical progression in SCSI's more than 20-year run is Serial SCSI, a low cost, high-speed serial interface. What remains to be seen is whether Serial SCSI can become a cost effective alternative to Parallel SCSI. In the meantime, Serial SCSI will compete with Serial ATA at the low end for disk within the desktop or workstation, as well as with Serial ATA and Fibre Channel at the backend interface for disk arrays.

Fibre Channel Fibre Channel will continue to erode SCSI library attachment as mid-range companies begin to share their backup resources through networking technologies such as Fibre Channel SAN. However, because Fibre Channel disks command a 400 to 500 percent premium over lower cost disk interfaces, Fibre Channel could lose the backend disk array bus market to ATA/IDE, Serial ATA and Serial SCSI. Overall, as Fibre Channel infrastructure companies continue to offer with low cost 2 Gb/second hardware, Fibre Channel will continue to be a dominant front-end interface over the next two to three years.

InfiniBand The InfiniBand fabric uses a switched fabric form of topology similar to Fibre Channel, and has three key components - Host Channel Adapter (HCA), Switch, and Target Channel Adapter (TCA). Several different companies offer each, and rudimentary interoperability testing has begun.

• HCA connects a host system to the InfiniBand fabric.
• The InfiniBand Switch interconnects HCAs to HCAs or TCAs, using information contained within the InfiniBand transmission that determines the destination port to direct the serial stream.
• TCA connects non-initiating devices to the InfiniBand fabric.

InfiniBand has significant potential to be the interface choice for blade level server clustering. Low latency, a simplified serial interface, and well-defined management provide blade server designers a fabric on which to build a tightly coupled parallel processing system. However, before trying to grow to the next level of data center fabric, InfiniBand must gain acceptance as the de-facto interface for clustering.

GigE (Ethernet plumbing supporting NDMP, iSCSI, FCIP & iFCP) Gigabit Ethernet (GigE) will have a profound effect on storage going forward because it provides the hardware and Ethernet TCP/IP foundation for several higher level protocols associated with data movement and backup, such as the Network Data Management Protocol (NDMP) protocol iSCSI, FCIP and iFCP protocols.

The following is a brief overview of a GigE system and its components, illustrating key building blocks. All of the current proposed storage over IP standards (iSCSI, FCIP, iFCP, NDMP) share a common hardware platform in addition to TCP/IP as a low-level protocol. For Hosts and Targets to communicate with each other, they must speak the same language. In other words, iSCSI hosts talk to iSCSI targets, while NDMP servers talk to other NDMP servers over GigE.

NIC - Network Interface Card - Provides the hardware bridge between the host's native bus and the 100/1000 Base-T Ethernet serial interface.

Switch - Provides routing between ports on the switch based on IP addresses contained within the TCP IP packet.

Cabling - GigE provides for several different media types depending on the required distance. Simple CAT 5 E & CAT 6 copper cabling will provide up to 100 meters from port to port. Short wave lasers provide up to 500 meters, while various long-wave lasers provide more than 80,000 meters.

Hosts / Servers / Appliances - For any of these systems or subsystems to take advantage of the Ethernet hardware and TCP/IP transport layer, they require a higher level protocol layer to define intra-packet formats and bit usage, such as iSCSI, iFCP, FCIP and NDMP.

One key advantage that 1 GigE has over previous generations of Ethernet (10 and 100Base-T) is that it permits only switched traffic flow. Until recently, most 10/100 LANs deployed were of the hubbed variety, resulting in packet collision. That was followed by re-transmission during times of heavy LAN traffic, making them less than ideal from a storage perspective. GigE does not permit hubs and has eliminated delays caused by collision, yielding a more predictable throughput than its hubbed predecessors.

NDMP Network Data Management Protocol (NDMP) is an established market for backup of high-end NAS filers. Originated by Bud Tools and popularized by Network Appliance as a means to prevent the need for third party software, NDMP is a higher-level protocol for Ethernet.

A tape library that supports NDMP version 4 enables data to be transferred directly from the NDMP-compliant NAS device to the library under the control of the backup server running NDMP client software.

iSCSI Internet SCSI (iSCSI) is a critical emerging protocol for Ethernet that will provide a means for block level delivery over the historically file-orientated Ethernet interface. Tape automation and iSCSI provide a clear winner for the low-end to mid-range customers for its ability to cost effectively share a tape library and drives across a low cost GigE LAN to help resolve the ongoing data backup crunch. Now tape drives with their 11-30 MB/s data rates will appear as locally attached drives to physically disperse servers within the data center at 1/6 to 1/10 of the cost of Fibre Channel.

iSCSI will replace serial SCSI as the workhorse interface of the IT department. This will enable network engineers to quickly become iSCSI SAN storage experts to help their company deal cost effectively with their growing storage demands.

FCIP Fibre Channel (encapsulated within) IP (FCIP) provides connectivity between islands of Fibre Channel SAN across the MAN WAN using TCP/IP. FCIP relies upon standard GigE hardware for switch functionality and routing through the IP LAN/MAN/WAN. FCIP bridges encapsulate incoming Fibre Channel streams from Fibre Channel switches and provide for connectivity within the data center.

FCIP devices should be thought of as simple protocol bridges between Fibre Channel and GigE, providing only the necessary hardware and firmware to encapsulate Fibre Channel into TCP/IP packets and vice versa. MAN / WAN link speeds can be low and latencies can be high requiring timeout adjustments in both SAN islands to allow for operation.

iFCP Internet Fibre Channel Protocol (iFCP) provides SAN-like fabric functionality based on Ethernet switching. It is a TCP/IP based protocol allowing a SAN to be created with Fibre Channel end devices using all GigE switching and routing in between. iFCP devices or "Gateways" devices provide both switching and bridging functionality. iFCP's main competition is from the iSCSI market, and the main vendor for iFCP, Nishan Systems, has addressed this by incorporating iSCSI ports on their iFCP switch. The following is an iFCP IP Fabric example.

10 GigE In June 2002, the IEEE standards body ratified 802.3ae, the 10 Gigabit Ethernet (10 GigE) standard, making 10 GigE the backbone or main pipe of choice for large data centers by mid 2004. Smaller data centers will adopt this protocol as costs to implement it decrease.

10 GigE also will become the point of complete hardware convergence between block level SANs and file level NAS as 10 GigE pipes concurrently support iSCSI, iFCP, NDMP, as well the file level-like NFS and CFS. Products from Cisco and other high-end networking vendors use 10 GigE as a switch-to-switch pipe. The cost for a single 10 GigE port is more than $10,000.

As costs for 10 GigE ports drop and 10 GigE switches become available, data centers will begin to deploy 10 GigE fabric capable of 1 GB/second as their "corporate" backbone. Over time, GigE will move to the desktop providing 100 MB/second bandwidth on the corperate LAN. Last mile costs should decrease then, providing 100 Mb/second access also across the MAN / WAN.

A host of interfaces exist for tape automation and continue to evolve in response to the dynamic needs of storage. For example, for the standalone drive and loader markets, tried and true parallel LVD SCSI is ideal. Enabling small- to mid-sized companies to create dedicated backup SANs is GigE. And for midrange companies seeking simple solutions to manage rapidly growing data, GigE NAS will continue to be robust.

So the next time company executives are "in your face" about your interface strategy, you can assure them that certain interfaces are best suited to work with your backup system. Whether InfiniBand, SCSI, Fibre Channel or GigE, an appropriate interface strategy is key to advancing your organization's storage system.

• Lee W. Payne is the Senior System Technology Strategist for Overland