Put the "Server" into "Storage Area Network"
By Brenda Christensen
Brocade Communications Systems Inc.
Before the Fibre Channel community started using the SAN acronym for STORAGE- area network, systems vendors used SAN to represent SYSTEM-area networks. Unfortunately, as the popularity of the term storage-area network has increased, vendors have often neglected to include server/host issues in their definition and discussions of SANs. It`s because of this critical link--the link between servers and storage--that I have continued to use the term server-storage network in lieu of storage-area network.
The trade press, analysts, and vendors are all comfortable using SAN to represent the emergence of the storage-area network as it completes the LAN/WAN circle. I, too, would adopt this acronym if I was certain that the industry not only included servers in its description of SANs, but more importantly was addressing server/ host issues during their design, testing, and implementation of SANs. However, I am not convinced of this fact.
The true deployment of Fibre Channel calls for products to be designed from an end-to-end solution perspective. Today, Fibre Channel products come from independent storage companies and storage divisions of system houses. However, for users to take complete advantage of Fibre Channel capabilities, the server and operating systems people and divisions of these companies must be closely involved in the development of SANs.
True, the initial adoption of Fibre Channel began at the disk-drive level, but products that enable the development of a Fibre Channel network (i.e., switches) are available, allowing end-users to create a network infrastructure between servers and storage. Now, we need to quickly change our way of thinking and communicating about Fibre Channel--to a system-level view that depicts an end-to-end solution. This orientation is essential if we are to meet end-users` expectations for robust, resilient, and manageable Fibre Channel SANs.
There is no question that servers should be computational engines and not bottlenecks. And there is no question that considerable talent and energy is being invested in the concept of connecting storage devices directly to networks to avoid the server bottleneck.
But whether the server OEM and Fibre Channel communities are investing enough resources in server I/O, operating systems, middleware, global file systems, and other server-based functions is uncertain. These design and support functions of the server and host-bus environments are critical for building a SAN in order to support two kinds of traffic: server-to-server (low latency and small messages) and server-to-storage (high throughput and large packets).
There is no magic wand that will incorporate all of the changes that need to be made to operating systems, I/O channels, and middleware so that end-users will be able to cluster servers and share storage. While OEMs today offer homogeneous solutions, information-intensive end-users live in a heterogeneous environment and expect SANs, too, to be comprised of heterogeneous systems. Users expect vendors to be unified in their requests to suppliers of operating systems, databases, and middleware so that heterogeneous SANs become a reality.
In developing Fibre Channel products, the host-bus adapter (HBA) has taken the most time to develop and has run into the greatest deployment difficulties due to its dependency on I/O architecture and operating system device drivers. Today, operating systems have a single I/O path, designed for legacy point-to-point connections, whereas networked storage requires point-to-multipoint communication. Fibre Channel`s ability to support concurrent storage and networking protocols on a single HBA is unique, but the operating system must also recognize multiple protocols. For business-critical applications, Fibre Channel HBAs should have dual paths for redundancy and availability so that the host can support failover mechanisms.
In addition, today`s host bus limits Fibre Channel`s performance possibilities. The de facto standard, 32-bit PCI, caps the I/O stream required for taking advantage of gigabit networks. Information-intensive companies are beginning to realize that they must demand servers that incorporate 64-bit PCI or next-generation I/O solutions. Also, to avoid overwriting disks and to prevent data corruption through multiple server access, database and operating system providers need to offer distributed lock management functions.
Many OEMs today also assume their customers will use only one operating system (theirs). And even if the systems vendor offers dual operating systems, the vendor still presumes customers will use these products in a homogeneous fashion. Today, enterprises share storage among multiple heterogeneous groups consisting of different servers and operating systems. Fibre Channel networks must provide the logical separation between these groups.
Finally, high on users` wish lists are network management features that provide vital information about all devices in the SAN. This information comes from the server environment and storage devices to the network fabric and must be represented as one logical network view.
Industry analysts such as Giga Information Group`s Anders Lofgren are correct in recommending Fibre Channel for use in homogeneous environments today, mostly because the functions required for sustaining heterogeneous environments are not yet available. To date, the priority has been resolving the problems of deploying physical devices, such as disk drives and storage arrays. This task has been accomplished. Now it`s time to address the lack of a global file system, distributed lock management, and tight integration of storage and servers and to overcome operating system limitations that will hinder the deployment of heterogeneous SANs.
I concede to the use of the SAN acronym and have added it to my vocabulary. However, with the publication of this article, I am launching a formal campaign to make server and operating systems issues a priority in all my communications and am making a formal challenge to all Fibre Channel vendors to join me in my efforts. The computer industry must fully engage in overcoming the limitations of today`s servers, operating systems, distributed lock managers, global file systems, and I/O channels for storage-area networks that meet tomorrow`s application requirements. Creating reliable and high-performance SANs that generate significant application benefits for end-users in heterogeneous environments will only come from these changes.
Brenda Christensen is vice president of marketing for Brocade Communications Systems Inc. in Santa Clara, CA.