It's a familiar story. As one part of a computing environment improves, others are forced to play catch-up. So progress in multimedia computing, data warehousing and other data-intensive technologies spurred advances in storage systems and local-area networks. Storage capacities are now doubling every 12 to 18 months and high-speed LANs are becoming commonplace. And those developments, in turn, have put new pressure on another part of the system--the storage interconnect. SCSI, the long-time workhorse of storage interconnect technology, is fifteen years old, and it's showing its age. Upgrades such as Fast/Wide SCSI and UltraSCSI I and II have helped squeeze more life out of the technology, but it's clear that in heavy-duty storage systems such as data center arrays, SCSI's days are numbered. Despite the improvements, its performance, availability and other features are falling further and further behind today's business-critical application requirements.
Inevitably, new-generation technology has stepped up to fill the gap. That technology is Fibre Channel Arbitrated Loop (FC-AL). Developed and standardized over the last few years, Fibre Channel interconnect technology improves dramatically over SCSI and other earlier-generation interfaces (such as IBM's ESCON interface) in performance, availability and scalability.
Fibre Channel's innate superiority and its widespread adoption by manufacturers--with a flurry of product introductions in 1997--will help it capture an ever-growing share of the array market over the next few years. A wide range of computing environments, from heavy-duty OLTP systems to data warehouses and Web servers, will benefit from this shift.
The benefits include Fibre Channel's greater performance, which translates to better scalability and flexibility; its high availability features; its support of switches (a critical advantage in cluster configurations); its longer interconnect distances; and its much-enhanced reliability. And, for all its advantages, Fibre Channel adds relatively little to the cost of a disk array.
What does this mean to you? In the short term, Fibre Channel simply offers a more robust and flexible way to build high performance storage systems.
Longer term, Fibre Channel represents much more than a vastly improved SCSI bus. It is the first interconnect to incorporate network-style technology into storage systems. For example, Fibre Channel supports connectivity increasing hubs and switches, similar to the hubs and switches used in Ethernet. This fundamental change in storage architecture has far-reaching implications for how storage systems get built and managed, as well as for the storage industry itself.
Soaring bandwidth requirements are quickly outpacing the capabilities of SCSI arrays. For example, consider an OLTP system that has been designed for retrieval of simple text-based records, consuming 2KB per record. At 4000 I/Os per second (IOPS), the total bandwidth needed is 8MB/sec.
That's easily within the range of a SCSI array. But suppose, as is often the case these days, the OLTP system is then upgraded to support multimedia content, as when images are added to an on-line product catalog. As a result, the size of each transaction jumps to 16KB, calling for an aggregate bandwidth of 64MB/sec--well beyond the transfer rates of Fast/Wide SCSI (20MB/sec) and even UltraSCSI (40 MB/sec).
With SCSI, there's little choice but to buy one or more additional arrays--at a typical cost of $50-75,000 per array--to handle the greater bandwidth requirements. Not so with Fibre Channel, which supports data transfer rates of 100MB/sec (full duplex). That's sufficiently high that a single array could handle the increased data load--no need to go shopping for extra storage systems.
Fibre Channel's greater bandwidth also provides a marked performance advantage in systems where the I/O volume is smaller but the files are really large. Take the example of a large data warehouse, where complex queries may require the writing of tables totaling 100 gigabytes or more. At 100MB/sec, a single Fibre Channel interface will write the file at more than twice the speed of an UltraSCSI connection. Even higher speed is possible by striping across multiple Fibre Channel connections, raising the data rate to gigabytes per second.
All of the performance in the world is useless if the array is down. Here is where Fibre Channel really shines. For starters, Fibre Channel disk drives are dual-ported, so an array can have two paths to each of its disks. This provides redundancy in case one link should fail, a vital concern in high-availability environments such as financial, manufacturing and electronic commerce systems. Nor are the redundant channels just hot standbys. Loads can be dynamically balanced across channels, increasing array performance and reducing I/O latency. Be aware, however, that not all of the new "Fibre Channel" arrays use Fibre Channel disk drives, so they cannot implement these features.
Moreover, Fibre Channel was designed from the ground up for high reliability, in marked contrast to SCSI, long known for its unpredictability. For example, Fibre Channel's drive hot swap capability is much more robust than SCSIs, assuring data integrity during drive removal and insertion. Unlike the case with SCSI, I/O operations don't have to be halted to do a hot swap. And Fibre Channel covers much greater distances than SCSI, allowing connections of up to 10 kilometers, versus 12.5 meters for UltraSCSI. This enables mirroring of data on separate premises for disaster recovery purposes.
Fibre Channel provides a significant boost in disk connectivity, one of SCSIs many shortcomings. Even though Wide SCSI is technically able to support 15 disks per bus, practical limits are lower. Most SCSI arrays top out at around 20 to 42 disks.
That limit makes SCSI arrays rather inflexible when it comes to expansion. Consider the example of an OLTP system designed to handle a peak of 4000 IOPS. That's pushing the limits of a 42-disk array which, with a maximum of 100 I/Os per second per disk, can handle perhaps 4200 IOPS. So what happens when growth in usage raises the peak load to 5000 IOPS (not uncommon in today's industrial-strength environments)? With SCSI, there's no choice but to add another array at considerable cost.
Wouldn't it be nice if the array could accommodate more disks? Capacity could be expanded from 4200 to 5000 IOPS simply by adding eight new drives to the array, a significantly cheaper alternative to buying a whole new array.
That scenario becomes possible with Fibre Channel. With its much larger addressing range, Fibre Channel supports 100 or more disks (126 theoretically) per loop. With many loops per system, that allows arrays to hold hundreds or thousands of disks, leaving ample headroom for future growth. Buyers can start small, with just a few disks, and scale up incrementally and economically to larger and larger configurations.
Unlike SCSI, Fibre Channel interfaces are built to work in conjunction with switches, also called a "fabric." Switching, widely used in networks but virtually unknown in storage, gives Fibre Channel a flexibility unknown in SCSI technology. For example, switches are a vital component in clusters where multiple hosts are attached to multiple arrays. The switch performs a function analogous to that in switched Ethernet networks, boosting bandwidth and the number of nodes in a loop. Such cluster arrangements are difficult or impossible to implement with SCSI, which has no provisions for switching.
Switches also can be used to tune the storage system to application needs. For example, a big data warehouse might use switches to provide large amounts of bandwidth, while an OLTP system might use them to attach many disks to the host (and boost IOPS).
Most purchasers of Fibre Channel storage will choose it strictly for the many strengths it offers today: performance, availability and connectivity. However, IT architects looking for competitive advantage see beyond these obvious benefits by looking at the impact networks had on information systems over the last decade. In this view, Fibre Channel is a platform for building the storage systems needed by our increasingly data intensive organizations. The potential impact of Fibre Channel on storage can be compared to that of Ethernet on networking.
Before Ethernet came out, networking was dominated by incompatible proprietary systems that were slow, inflexible and difficult to manage. When Ethernet first came out people liked it because it was fast, flexible and standards-based. From simple beginnings, dozens of companies jumped in to add functionality (repeaters, routers, hubs, bridges, switches, print servers and more), enabling a powerful new computer paradigm whose effects are still unfolding and creating a multi-billion dollar industry.
Today's storage arrays and pre-Ethernet networks share important similarities. Arrays are proprietary, fixed architecture boxes: if you need more capability than is designed into the box (IOPS, bandwidth, host ports), you have one simple and expensive choice--buy another box. Innovation is slow: some of today's most heavily promoted arrays were designed almost a decade ago. Management is clumsy: each additional array is managed as an individual entity. The long-term promise of Fibre Channel is as simple, and as revolutionary, as Ethernet's promise was more than a decade ago: a fundamentally better way to store, access and manage large amounts of data.
Yet FC-AL is not a simple replication of existing network technology. Designed for storage applications, its inherent manageability is much better than data com networks. For example, FC-AL networks are small on scale, limited to 126 nodes (mostly disks), and largely self-managing. Larger FC-AL networks are built up from these small building blocks. Initialization is automated so there are no network addresses to set. Hubs are transparent to software and need no software configuration. Switches operate at the lowest levels of the protocol stack for operational simplicity. The upper level storage software protocol, SCSI- 3, is well optimized for storage applications. Most importantly, if you don't want to use Fibre Channel's added functionality, you don't have to. Fibre Channel can be used as nothing more than a faster and more available SCSI bus with its additional features hidden until you need them.
Fibre Channel is here. Standards are in place. More than 100 companies are building Fibre Channel products. With all its benefits, it is only a matter of time before Fibre Channel eclipses SCSI. Analysts say that time is soon. According to International Data Corp., Fibre Channel should dominate the mass storage market by 2001.
As you consider the benefits of Fibre Channel, look at how your server I/O profiles have changed in the last few years and how they might change in the future. Will new applications require much higher capacity? Bandwidth? I/Os? Is it difficult to predict requirements? Do users frequently demand new applications? If you answered "yes" to any of these questions, chances are you would benefit from Fibre Channel storage arrays.
In very short order, Fibre Channel stands to fundamentally alter the way companies choose, configure and manage their mass storage subsystems, and now is the time to begin planning for that transition. IT managers would do well to include Fibre Channel in their buying decisions today.
>Fibre Channel-Arbitrated Loops Enhance Data Warehousing
Many IT organizations are turning to Fibre Channel technology (which runs on top of the SCSI protocol) within storage area networks for a solution. Fibre Channel is a set of standards that enable greater connectivity, and rapid and reliable data transfer (100 MB/second) using the SCSI interface. A Fibre Channel-Arbitrated Loop (FC-AL) supercharges SCSI by allowing for a large number of attached devices (up to 127) with distances of up to 10 kilometers. This type of storage interconnect offers greater reliability, availability and scalability than on the current network storage. This is important for supporting data warehousing devices such as on-line analytical processing (OLAP) servers.
The other piece necessary for leveraging data warehousing and OLAP technologies enterprise-wide is a single-system view management tool, in addition to device level management. This advanced management occurs on two levels: managing so that IT can get a complete picture of the loop in order to isolate problems quickly and global management of all storage networks in an enterprise from a single console. Management requires advanced loop integrity features such as complete loop diagnostics, problem detection, isolation and recovery. Comprehensive loop management can virtually eliminate problem detection and isolation time--which can be up to 95 percent of the total outage. What we do know is that system administrators spend most of their time in a "fire- fighting" mode. There is simply no time for these people to sit in front of a console and look for failure mechanisms.
Vixel offers storage interconnect solutions using ANSI-standard FC-ALs with enterprise-level management. Its products offer a complete line of storage interconnect products including Fibre Channel switches, and managed and unmanaged hubs, that allow for its customers to economically build comprehensive storage network solutions. Its loop management software enables rapid problem detection and isolation, expediting recovery which allows system administrators to maintain a smooth running enterprise. As a leader in storage interconnect development, Vixel is also developing V2, an integration laboratory, designed to allow Fibre Channel vendors to test integrated solutions for all aspects of enterprise-wide storage networking.
The enterprise is at a crucial threshold--on the verge of unleashing the power of data-intensive applications and data warehousing. Storage network solutions using Fibre Channel hold the key to making this a reality.
Jeff Vogel is a vice president at Vixel Corporation, a leading developer of Fibre Channel storage interconnect solutions, including managed and unmanaged hubs, and optical transceivers. Vixel provides efficient and affordable storage interconnect solutions for data-intensive applications. For more information, visit www.vixel.com.
Register or login for access to this item and much more
All Information Management content is archived after seven days.
Community members receive:
- All recent and archived articles
- Conference offers and updates
- A full menu of enewsletter options
- Web seminars, white papers, ebooks
Already have an account? Log In
Don't have an account? Register for Free Unlimited Access