OEMs need flexibility, upgradeability when integrating Fibre Channel support: understand the tradeoffs, know the options.

With demands for networked storage growing, the choice for OEMs isn't whether to incorporate Fibre Channel storage connectivity in servers, but in what form to provide that support. The choices range from very tight integration on the server or storage system motherboard, to quasi-custom integration through "mezzanine" boards for a blade server (or other custom applications), to the most flexible option of providing expansion slots for Host Bus Adapters (HBAs).

OEMs should choose different options for different customer sets at different points during the adoption curve of any given I/O (input-output) standard such as Fibre Channel. Among the variables to consider is the adoption rate of the I/O technology, how quickly that technology is changing, and how much flexibility the customer needs in changing or upgrading their storage interconnect.

Regardless of the form factor, OEMs should choose the I/O technology from a vendor who offers the greatest flexibility for OEMs and customers, the greatest ease in upgrading their technology, the largest installed base and the highest brand awareness in the marketplace.

The Fibre Channel Challenge

Organizations of all sizes need to constantly store more data due to an increased reliance on information systems, the data generated by 24x7 e-commerce, as well as new regulations that require organizations to keep more data for longer periods of time.

The form factor question has become all the more compelling as a new class of cost-conscious FC users emerges, namely small and medium business (SMBs). These users currently employ DAS and face issues very similar to those that first led enterprises to turn to networked storage. This new market segment for Fibre Channel broadens the discussion to the low-cost volume server market.

In the past, customers might have met these needs by adding more direct-attached storage to individual servers. However, as the number of servers grows, this becomes increasingly hard to manage and inefficient because customers might be forced to add new disk hardware to one over-taxed server while storage sits nearly empty on a nearby, underused server.

Fibre Channel has been the protocol of choice for high-end storage needs because of its high throughput, reliability, scalability and the maturity of the applications used to manage and administer it. FC offers three times the throughput of SCSI (Small Computer System Interface), improving overall data throughput and reducing the "backup windows" during which applications must be taken off-line to allow data to be backed up. Fibre Channel also allows servers to be connected to storage devices over distances of as much as six miles, making it easier to perform remote data backups to provide business continuance in the event of a disaster at the main data center.

For many years, large enterprises have used FC storage-area networks (SANs) to aggregate their storage. By allowing servers to share a single pool of networked storage, they allow the more efficient use of all available storage, improve the availability of data, and reduce network congestion by removing backup traffic from the corporate data network.

More and more organizations in the SMB markets are also looking at SANs built around lower-cost disk drives. While these arrays may not offer the same levels of scalability, throughput and reliability as Fibre Channel technology, they meet the requirements of many SMB customers and are an increasingly cost-effective alternative to direct-attached storage. Finally, the increasing processing power of even industry-standard, Windows-based servers makes them able to run applications, which require the block-level data access provided by Fibre Channel.

However, different customers might need different forms of FC support depending on their price sensitivity, the amount of physical space available for their servers, and how often they'll need to upgrade their infrastructure. OEMs don't want to ship server hardware with more I/O capabilities (and thus more cost) than a customer is willing to pay for, or to ship I/O capabilities that won't work for more than a year or two due to rapidly evolving storage standards. OEMs can, however, capture market share by shipping the right Fibre Channel support in the right form factor to the right customers for the targeted market.

To understand the trade-offs, it's important to understand what constitutes Fibre Channel support. The first requirement is a physical FC connection, which supports different types of cable (optical as well as copper) as well as the different types of Fibre Channel connections. The two key connection methods are loop mode, which allows as many as 126 devices to share the connection directly in a loop, or a switched fabric that allows thousands of devices to share exclusive point-to-point connections.

Fibre Channel also requires hardware and software to support the proper formatting of data packets and other functions such as controlling the flow of the packets, correcting any errors in transmission, and (in the case of TCP/IP) reassembling packets in the proper order when they arrive at the server.

To provide the highest possible server performance, the vendor may want to offload as much I/O processing as possible from the server CPU to the HBA. In Fibre Channel, this can be accomplished by adding a specialized I/O processor with the associated additional memory required for data buffers.

Which Form Factor? The Choices

Each form factor for FC support offers different costs and benefits.

An embedded approach to Fibre Channel support takes up the least space and reduces up-front manufacturing and inventory complexity for the vendor since it reduces the number of SKUs (stock keeping units) the vendor builds. On the other hand, with support designed-in on the motherboard, every server ships with the cost of Fibre Channel regardless of whether every customer needs it. If choosing to "hard wire" Fibre Channel, an OEM should look for a vendor whose technology allows upgrades by replacing the firmware rather than replacing the I/O processor or other physical components.

A vendor producing small "blade" servers can put the FC components on a mezzanine card attached to the blade server. While such mezzanine cards adhere to the Fibre Channel standard in areas such as electrical signaling and data formatting, they often have a physical form factor unique to each server vendor. Such cards require little extra space in the server, and are far easier to upgrade than components built onto the motherboard. They do, however, require the customer to purchase semi-proprietary hardware, which can generally not be replaced by the customer. An OEM wishing to provide the greatest investment protection for their customers should, again, look for a vendor whose mezzanine cards can be updated with only a firmware upgrade rather than a physical replacement of the card.

The third, and currently most commonly used option is a HBA--an industry-standard add-in card plugged into a slot in a server or used as a target device for storage systems. This requires no extra effort for OEMs building full-size servers since those servers already ship with a complement of industry-standard slots. It also provides the maximum flexibility for customers, since they can easily add any level of FC support whenever they want it.

Choosing a Form Factor

In choosing which form factor is right for a given customer, an OEM must consider the available space, how quickly the I/O technology is being adopted, and how quickly that technology is changing.

In choosing a physical form factor for their FC support, OEMs need first to categorize the types of customers they serve, and the types of servers they produce for those customers. Are they providing Fibre Channel connectivity for full-sized servers used in an SMB environment, for example, or for a major corporation that needs small blade servers to reduce their physical data center space, as well as to cut their heating and cooling costs?

Next, they should consider how quickly or widely customers are adopting the I/O technology. The higher the adoption rate, the less risky it is for a vendor to embed support for the technology into every server or storage system it ships.

The final factor is how quickly the technology is changing, and how often customers will want to upgrade to new versions of the I/O standard. Early in the life of a standard, vendors will rapidly improve its performance or add new features, such as support for new devices like tape drives. If an I/O standard is still in a period of rapid change, customers will opt for a form factor that makes that technology easiest and least expensive to upgrade.

Embedded support for Fibre Channel, then, might be best for well-established, mature technologies with very high adoption rates. In such cases, few customers will mind paying an incremental cost to have the technology built into every system they buy, rather than use an expansion slot or buy a mezzanine card. However, it is critical in this situation to ensure flexibility by implementing solutions with an upgradeable, firmware-based architecture. Providing support through a mezzanine card might be the best option when the customer needs a blade server but still wants a choice of support for different types of storage devices. Using a traditional HBA is the most common connectivity method today and is the best approach when the technology is changing most quickly, where space is not critical, where adoption of a standard isn't yet universal and where customers want the maximum upgrade flexibility.

I/O Vendor Requirements

As a technology standard matures and customer needs change, the "right" form factor for that technology may change. But regardless of the form factor, OEMs and their customers need the utmost in flexibility and upgradeability.

Emulex provides the industry's most flexible upgrade paths through its Service Level Interface, which abstracts the I/O controller and its firmware from the driver stacks. This makes the driver hardware-independent, reducing the risk of obsolescence for an OEM embedding Emulex HBAs or I/O controllers into their products. This lowers the workload and risks for end users, as well as for vendors who can leverage the testing done on existing drivers rather than perform a full re-qualification of new drivers for next generation HBAs and I/O controllers. Emulex also offers customers the assurance of a brand name whose I/O technology is delivered by the world's leading server and storage OEMs.

Summary

Small footprints, high performance, maximum flexibility and support for industry standards are key purchase criteria for server and storage customers who demand high-performance block-level access at ever-lower price points. Fibre Channel connectivity support can be provided in three different form factors: embedded I/O controllers, on a "mezzanine" card on a blade server, or on a HBA in an industry-standard expansion slot.

Which form factor is right for a particular customer set depends on the maturity of the I/O technology, the adoption rate of that technology and the end customer's space requirements. Emulex allows OEMs to more easily satisfy these disparate needs because of its innovative driver architecture, which allows "software-only" upgrades to support new I/O performance levels and features and offers the broadest range of interoperability. Emulex's reputation as a premier source of intelligent building blocks for network storage solutions makes it the choice for vendors who need to provide flexible, cost-effective Fibre Channel support in server and storage hardware, regardless of the form factor.

www.emulex.com

Joe Teolis is senior director of product marketing at Emulex Corp. (Costa Mesa, CA)