Whither Fibre Channel?
iSCSI will move up the storage food chain, but nobody's ripping out their Fibre Channel storage anytime soon. (Originally published in IT Architect)
August 1, 2005
Promise: The FC and iSCSI industries both say their technology represents the future of storage--FC because it's fast and efficient, and iSCSI because it's cheap and easy.
Players: On the FC side, McData, QLogic, and other members of the Fibre Channel Industry Association (FCIA). On the iSCSI side, EqualLogic, Network Appliance, and members of the Storage Networking Industry Association (SNIA). Some vendors, such as Brocade, Cisco, and IBM, participate in both groups and offer both types of storage.
Prospects: FC has several good years ahead of it--probably five. During that time, iSCSI and InfiniBand will start to creep into the data center. Gateways are being built that will link FC to iSCSI technologies, and the platform of the future will most likely be a hybrid of the two.
Much of the storage buzz this past year has been about new iSCSI SAN products, which are doing brisk business in small and midsized companies that appreciate the lower cost and ease of IP storage networking. At the same time, however, Fibre Channel (FC) SANs are quietly improving performance and market share. This has spurred a lively debate in the storage industry: Are FC's days numbered, and will iSCSI take over?
IS FC FRAYING?
One school of thought is that FC will increasingly be superseded by iSCSI. Although proponents concede that there's still a need for 4Gbps FC today, they insist there won't be two years from now when 10 Gigabit Ethernet comes to storage networks. "Over time, people who are backing FC technologies will have to spend more time proving why FC is superior to Ethernet, given that they're going to have comparable performance and Ethernet is going to have a lot of advantages when it comes to volume and commitment to building lots of systems," says IDC analyst Rich Villars.
Generally speaking, iSCSI is easier to use--at least for IT architects familiar with Ethernet and IP--and more affordable. While most iSCSI storage currently uses a 1Gbps Ethernet link (FC implementations run in 1, 2, and 4Gbps modes), by using Microsoft Multipath I/O it's possible to aggregate bandwidth by running multiple Ethernet lines and load balancing between them.
iSCSI's lower cost is probably its greatest virtue. One company that was buying hundreds of terabytes of FC storage simultaneously bought an iSCSI storage system for its Microsoft Exchange servers. It turned out that new FC host bus adapters for the servers would cost more than the Exchange servers themselves, yet each one had a built-in Ethernet connection.
Security services and schemes for address takeover that facilitate fault tolerance already exist in the IP world. iSCSI also inherits IP interoperability rules, whereas with FC there are still occasional compatibility issues between adapters, switches, and storage devices made by different vendors.
iSCSI is catching on in the SANs of small and midsized businesses, especially in Windows and Blade environments where performance isn't as critical and it's hard to cost-justify hiring an FC storage specialist. It's also likely to creep up into higher-end uses. "History is on the side of iSCSI on this," says Clod Barrera, director of systems strategy for IBM's Storage Systems Division. IBM offers both FC and iSCSI storage systems. "The track record of technology in general is that low-end technologies tend to grow upward in scope and reach, and high-end technologies almost never grow downward. It's plausible that iSCSI works its way up into higher-capacity, higher-performance commercial computing, and it's less plausible that FC works its way down."
Of course, many high-end technologies have grown down, including Unix, Windows NT, Gigabit Ethernet, and computers themselves.
LONG LIVE FCFC vendors contend that their technology will remain dominant in high-performance networks, the midmarket, and the enterprise market--at least throughout this decade. "There's still a huge installed base of FC supporting larger enterprises and applications that's not going away," says IDC's Villars. Where today approximately 40 percent of servers are attached to FC, analysts predict that more than 70 percent of servers will be attached to FC by 2007. (However, iSCSI vendors counter that this is merely a result of existing customers' growing data volumes, not new buyers.) iSCSI's server penetration is expected to grow from 3 percent to 12 percent in that same time frame, although IDC predicts that in 2009, 28 percent of all terabytes shipped will be iSCSI, up from a few percentage points today.
FC's strengths are performance, speed, and efficiency. "There's nothing available today that competes with FC at high speed and high capacity," says IBM's Barrera. FC uses a relatively lightweight protocol that makes minimal demands on servers, whereas iSCSI is more complicated--iSCSI on top of TCP on top of IP, with checksumming going on at various layers in the stack. (As silicon gets cheaper and as more of the TCP/IP stack is implemented in hardware, that advantage will diminish.) For high-speed, machine room-class secure enterprise storage and for demanding computer work such as database transactions and video or audio streaming, FC is the answer.
For most users, the most important measure of performance is I/Os per second--in other words, how fast the system can perform reads and writes to and from the storage system. The tricky part is that what vendors can achieve under optimal conditions doesn't always match what organizations will get in their data center. However, the vendors' numbers are a starting point.
In April, behind-the-scenes storage technology developer Engenio announced a 4Gbps FC storage system offering 550,000 I/Os per second burst from cache and 79,000 I/Os per second sustained from FC media. (One of the fastest iSCSI products, EqualLogic's PS Series, is said to achieve 360,000 I/Os per second.) IBM licensed the technology in its TotalStorage DS4800 disk system in June, and QLogic and Brocade announced compatible switch and host bus adapter products around the same time, making an end-to-end 4Gbps solution possible. Cisco Systems, McData, StorageTek, NCR, and others are sure to follow. The vendors insist that 8Gbps FC will be broadly available in two years, and that these higher-performing FC products will be backward compatible with existing 1 and 2Gbps FC infrastructures.
Barrera points out that FC is well-established where efficiency and physical security matter. "Not only do customers use it, but software stacks know how to manage it and operational processes use existing tools to install hardware, observe performance, and make changes where necessary," he says.Even iSCSI vendors say niche markets for FC may persist for a very long time in areas such as broadcasting, digital imaging and editing, and online transaction processing.
INFINIBAND OPTION
Another connectivity option, InfiniBand, has found a niche in high-performance computing environments such as Oracle and the trading floors of brokerage firms. Its logical use is in situations where companies use InfiniBand switches for clustering. Engenio president and CEO Thomas Georgens says InfiniBand doesn't pose an immediate threat to FC, but that its day is coming. While the company's next round of products will support InfiniBand and FC, he notes that InfiniBand has a narrow market, one skewed toward Linux.
PEACEFUL COEXISTENCE
The answer to the FC vs. iSCSI question most likely isn't one or the other, but a combination or hybrid. There's room for iSCSI, FC, and InfiniBand to each serve certain storage needs, and all camps agree that people who buy storage are slow to rip out and replace existing infrastructure. There will be a period of coexistence ranging from two years to eternity, depending on whom you ask."The storage world is a risk-averse world," says Barrera. "Nobody ever got fired in storage for being a little late on the coolest technology. What you get fired for in storage is having unavailable data or putting together an unreliable or twitchy infrastructure. Any movement away from FC is going to take time."
Barrera predicts that in five years, we'll start to see virtual FC circuits running over IP networks. Some underlying security and virtualization services will be provided by the IP network (on 10, 20, 40, or 100Gbps Ethernet), but management, performance, provisioning, and other upper-layer functions will be provided by software that thinks it's working with an FC infrastructure. Reads and writes would flow over FC protocols, but the physical infrastructure could be almost anything. "If I were a data center manager, I'd plan my roadmap for the next four years in FC and work to understand what gateway technologies exist to let me link iSCSI environments to the machine room," he says.
Senior Technology Editor Penny Lunt Crosman can be reached at [email protected].
You May Also Like