Written by Ryo Koyama|
Reprinted with permission from author
By now, you've undoubtedly heard of the Internet Small Computer System Interface (iSCSI), the emerging protocol standard that allows SCSI devices to interconnect over TCP/IP. If you think that iSCSI is an evolutionary technology that provides Ethernet ubiquity to storage area networks (SANs) and block capability to network attached storage (NAS), you are correct on both counts. What's even more exciting, however, is the revolutionary impact that iSCSI will have on Gigabit Ethernet and how this will forever change the direct attached storage (DAS) environment.
The combined effects of improved storage technology and the accelerated expansion of the Internet have fundamentally changed the way storage is used. Not so long ago, we struggled with files larger than 1.44 megabytes. Today, we access hundreds, even thousands, of megabytes from our cell phones, PDAs and portable music players. For the consumer, graphics and audio files (.jpg, .mpg, .mp3) take up hundreds of gigabytes of our personal storage; in the office, presentations, product documents (.ppt, .doc, .pdf) and an ever-increasing number of e-mail files (.pst) do the same. Add to that the expectation that these bigger-than-ever files be instantly and forever accessible, and the pressure to expand storage capacity is enormous.
In the minds of CIOs and IT managers everywhere, the role of storage has clearly shifted from tactical to strategic. Because of the growing sophistication of storage managers today, simply adding one more disk to a server just won't cut it. Since the peak of the Internet boom in January 2000, the storage networking market has grown from a $1.3 billion dollar industry to an estimated $4.7 billion dollar industry in 2003. For the years 2000 through 2004, the total storage market is expected to exceed $25 billion dollars.
As the name implies, DAS is storage that is connected directly to a computer. Typically, a hard disk or JBOD connected via a SCSI or IDE bus moves blocks of data via low-level commands. DAS is still the most common way to add storage to a network, but the inherent limitations of this technology preclude its ability to address the ever-increasing user demand on storage:
The major downside of DAS is that, unlike NAS, it must be accessed through the server on which it is attached, using valuable resources of the host machine and impeding direct assess to the storage within a SAN environment. However, there is now an upside in solving this dilemma by leveraging the combination of iSCSI, transport offload, and Dynamic Storage (available in Microsoft Windows XP and 2000) companies can now transform their existing DAS into a robust storage networking solution.
In principle, Gigabit Ethernet provides throughput equivalent to that of a 32-bit 66MHz bus. Unfortunately, because of computational limitations, these performance levels are seldom reached; and if achieved at all, come at the expense of overall system performance. Ethernet relies on the Transmission Control Protocol and the Internet Protocol -- collectively known as TCP/IP -- to ensure data integrity and to provide its robustness. True, TCP/IP's success in scaling with the ever-expanding Internet is a testament to its staying power, but this capability comes at a significant cost -- computational load. Industry experts agree that roughly 1 MHz of modern processor operation is needed to move a single megabit of TCP/IP data. Within any high-speed networking environment, then, processing the TCP/IP protocol stack to achieve full-duplex Gigabit Ethernet will fully consume a 2GHz processor, leaving no headroom for applications or other functions.
To address this limitation, an industry-wide initiative developed the transport offload computing model. It differs from legacy computing architectures by shifting responsibility for protocol processing from host to network interface. As a result, the network interface performs some or all of the functions corresponding to layers 1 through 5 of the OSI reference model.
To support this new computing model, various vendors are now producing transport offload engines (TOEs), dedicated semiconductors that perform high-speed TCP/IP computation. Additionally, TOEs rendered in dedicated logic provide low-latency throughput, effectively transforming Ethernet from a simple networking connection technology into a long-haul, high-speed interconnect bus.
iSCSI is a ratified IETF (Internet Engineering Task Force) standard for encapsulating SCSI commands over TCP/IP. This allows I/O block data to be transported over Ethernet networks in the same way that FibreChannel provides the functionality for SANs. The real beauty of iSCSI is that, since it operates on top of pervasive Ethernet networks, it can easily be added to existing networks, without requiring any new infrastructure.
For example, imagine you want to add disk storage to your exchange server. In the old, pre-iSCSI days, you had two common choices: Take the machine down, or have a SAN service your exchange needs. The second option is quite expensive unless, of course, you already have a SAN available. With iSCSI, you can simply assign storage from anywhere on your network and make it appear to an exchange server as if it is a local, physical drive. Mail stays up, storage is ultimately scalable, and you don't have to touch your existing network infrastructure.
Unlike Fibre Channel, iSCSI also provides system-level elegance by requiring no separate network adapter. Because the protocol is built on top of Ethernet, the same adapter can service both networking and storage traffic simultaneously. In addition, since the iSCSI standard calls for Internet Protocol Security (IPSec), the card will likely bring that capability to your system as well. Credit Suisse First Boston predicts that -- as IT managers begin replacing their 10/100 network interface cards (NICs) with 10/100/1000 NICs -- total Ethernet deployment will increase from 2.6 million units in 2002 to 20.2 million units by 2006, driving a compound annual growth rate (CAGR) of 67.2 percent. By choosing an iSCSI-capable Gigabit Ethernet adapter, system managers will derive the benefits of both higher throughput and iSCSI storage networking from a single adapter.
Today, a growing consensus acknowledges the need for gigabit throughput (versus 100 megabit throughput), especially as it relates to the user at the desktop level. With the ever-expanding need for user storage, iSCSI may very well be the killer app for Gigabit Storage.
iSCSI and TOE technologies allow DAS to become immediately networkable, but that alone does not ensure that it will be truly usable and, more importantly, widely-deployable. Over the years, the role of storage has changed. A decade ago, drive partitions made sense; today, a user's Inbox and MyDocuments folder are the most common means of segregating data. In the latest versions of its operations systems (Windows XP Professional and Windows 2000), Microsoft has introduced Dynamic Storage, which allows a disk to be initialized as a dynamic disk that can contain dynamic, simple, spanned, striped, mirrored and RAID-5 volumes.
The benefits from the end user perspective is that Dynamic Storage allows their storage to be infinitely scalable, whether for home or work, and accommodates the use pattern of today's computer user. From an IT manager's perspective, the compelling advantage comes from configuring all DAS (including desktop storage) as Dynamic Storage so that it all becomes immediately manageable and scalable.
In the same way that Windows 3.1 drove the mass proliferation of hardware acceleration in graphics cards, Dynamic Storage will drive the mass propagation of iSCSI and TOE in Gigabit Ethernet adapters.
Previously, I had the good fortune of working for the late Dr. Mark Weiser, who is regarded as the father of ubiquitous computing (www.ubiq.com/h ypertext/weiser/UbiHome.html). Weiser pioneered the idea that a technology, to become truly pervasive, had to be invisible to the user.
And, to be invisible, three vectors must intersect: (1) the use must be completely and intuitively understood; (2) enabling technology must be designed elegantly and perform the function without introducing new impediments; and finally, (3) collective solutions must be widely available so that access is not an impediment to its proliferation.
Users' ever-expanding needs for scalable storage solutions are clear, wellunderstood and urgent; iSCSI and Dynamic Storage exist and are designed to be easily deployable; Microsoft Windows is the most widely used operating system in the world and iSCSI and TOE capability is being built into industry standard Gigabit Ethernet cards.
The DAS industry is about to evolve in a big way.