Better grade for smart switches

06.03.2006
For many years, vendors have been promoting intelligent storage switches, particularly for virtualization projects. Initial implementations, however, relied mainly on appliances that created performance bottlenecks and were application-specific and difficult to scale. As a result, end users have been slow to adopt those so-called smarter solutions.

"I don't see any real intelligence yet," says Chris M. Christian, senior design lead SAN architect at Norfolk Southern Corp., a freight railway company in Norfork, Va. "Advanced intelligence features probably won't arrive till at least the summer, and maybe even later."

Though the products may not be ready quite yet, more intelligence has been added, and a lot more is on the immediate horizon. The driver? The frenzy surrounding information life-cycle management (ILM), which requires a rapid ramp-up in intelligence in order to move data around, regardless of the disk array, platform or servers. As a result, products are now coming onto the market that offer additional switch-based services and hold the promise of easing the storage management burden.

"Moving storage functionality into the switch offers the potential for long-overdue management relief when compared to traditional, host-based techniques," says Fred Moore, president of Horison Information Strategies, a storage consultancy in Boulder, Colo. "Many applications are ideal for this, including virtualization, volume management, backup, replication, storage resource management, hierarchical storage management and security."

Compared with traditional data communications switches and routers, storage switches have always had to provide additional functions that could be classified as intelligent.

Device discovery, for example, is essential for storage networking. Servers are initiators of storage transactions, but storage arrays are passive targets. For servers to find and establish connections with storage arrays, the fabric switch provides fabric log-on services and a simple name server. Once storage devices are registered with the fabric, servers may query the switch to discover potential targets. Likewise, fabric zoning and the mapping and masking of logical unit numbers are intelligent functions designed to ensure that only authorized devices are allowed to communicate across the fabric.

The current wave of intelligence, however, generally refers to higher-end switches known as directors. Today's director-class switches contain two types of intelligence: transport intelligence and application intelligence.

Transport intelligence provides enhancements for moving data through the storage fabric while optimizing throughput. The functions mentioned above can be classified as transport intelligence. Such features have been strengthened in the newest directors, which also include a new brand of transportation smarts, such as the ability to handle virtualization and the partitioning of resources.

Application intelligence, on the other hand, provides higher-level storage services, such as storage virtualization, continuous data protection, and backup and recovery.

There are only a few examples of application intelligence, but most intelligent-switch vendors are really promoting greater transport intelligence. For example, Broomfield, Colo.-based McData Corp. offers the Intrepid 10000 Director (i10K). Compared with the previous-generation directors, the i10K comes with advanced transport intelligence intended to speed up traffic.

Norfolk Southern is introducing an i10K into its environment. The current storage-area network (SAN) serves two data centers that represent 500-plus Unix servers (mainly AIX) and more than 600 Windows-based servers. The core of the SAN is built around four McData 6140 directors (two at each site) as well as a series of smaller McData 3232 switches at the edge of the SAN. It is also served by multiple tiers of data. An HDS 9980 array from Hitachi Data Systems Corp. serves Tier 1, an EMC Clariion CX700 from EMC Corp. is at Tier 2, and an EMC Centera is used for disk-based archiving.

Christian has redesigned the SAN with two i10Ks as the backbone. The other storage assets run through the McData director. Like many corporate users, though, the railroad remains cautious about adopting untried intelligence features.

"I'm buying the McData i10K to get I/O from here to here quickly, not for replication, backup or anything else," says Christian. "But it does take a lot of intelligence to move I/O efficiently in a SAN."

Packing the Trunk

In addition to boosting line speed, the latest brand of transportation intelligence includes trunking and path-selection features. For example, San Jose-based Brocade Communication Systems Inc. has made its Inter-Switch Link (ISL) Trunking software available as an optional add-on to its 4Gbit/sec. SilkWorm directors. ISLs that connect one switch to another can be combined to multiply the throughput -- i.e., four 2Gbit/sec. links become an 8Gbit/sec. line.

The San Diego Supercomputer Center (SDSC) at the University of California, San Diego, has deployed a Brocade SilkWorm 48000 director and plans to use this to implement trunking.

"With a 4Gbit/sec. switch, you can trunk together eight ports to get up to 32Gbit/sec.," says Bryan Banister, manager of storage systems and production servers at SDSC. "But we'll need to add extra 48000s to achieve this."

In the meantime, he has found a performance benefit of about 1Gbit/sec. using another intelligent transportation enhancement built into the director. The technology, known as Dynamic Path Selection, automatically routes data to the most efficient path available. This feature can also reduce the number of ISLs needed.

Other switch vendors are releasing similar products. McData switches, for instance, contain trunking technology, and Norfolk Southern has taken advantage of that feature to tie four of its 2Gbit/sec. McData 3232 switches into an 8Gbit/sec. trunk.

Still another largely transportation-based intelligence feature is known by a variety of terms, including virtual SAN (VSAN), dynamic partitioning or SAN routing. The idea is to split one SAN up into multiple discrete sections or tie several SANs into one central switch. This solves problems such as having low utilization on one tape system and overutilization on another.

With a VSAN, one switch -- such as a McData i10K -- can be partitioned into logical segments, and utilization rates can be optimized. Similarly, one part of the SAN can be earmarked to run at 4Gbit/sec. for backup while the rest runs at 2Gbit/sec. Management is also simplified, and access rights can be set for each partition.

"VSANs allow different fabrics to exist on the same switch, and each is unaware of the existence of other fabrics," says Moore.

Vendors are exporting this technology with switches that extend their traditional Fibre Channel-only range to other protocols, such as Fibre Channel over IP, Internet SCSI and Fiber Connectivity.

Cisco Systems Inc.'s MDS Multilayer Data Center SAN Switch is one example. Others include Hewlett-Packard Co.'s HP StorageWorks Multi-Protocol Router and the Brocade Multiprotocol Router (MPR).

America Online Inc. uses the Brocade MPR primarily for data migration and resource sharing across three SANs -- one for each data center in Northern Virginia. Two MPRs are stationed in each SAN, with a total of 2.5 petabytes stored.

Dan Pollack, operations architect at Dulles, Va.-based AOL, explains that when he used to migrate data from one site to another, it involved a lot of physical cabling, reconnecting and reconfiguring to establish dedicated lines for migration. Now he doesn't have to combine the fabrics to transfer data.

"We can provision in less than a day now versus a couple of weeks to physically set everything up before," says Pollack. "The built-in data-migration engines and data-processing engines of the Multi-protocol Router add more intelligence than the typical switch, as they allow us to move data around more effectively."

Pollack offers a virtual tape library (VTL) as an example. Instead of having to deploy one VTL per disk array or SAN, AOL can now share a VTL across arrays at all three sites. This has helped eliminate backup bottlenecks on large servers by moving traffic from Ethernet to Fibre Channel.

Application Intelligence

Much of the excitement around intelligent switches to date has centered on transportation. That's changing, though, as the demands of ILM force the integration of more application-rich features. The creation of multiple classes of data delivery and multiple zones of virtualized storage, for example, lets users deploy the SAN in alignment with business needs without having to add more hardware or software elements.

Take the case of storage security. Software-based encryption can exact a performance penalty, but deploying encryption appliances across a SAN doesn't always scale well and can be expensive. An intelligent switching platform, on the other hand, can add encryption across a SAN without encountering problems with scaling, and it can keep costs from escalating.

Another example is storage virtualization integration. Early attempts at virtualization largely entailed external appliances. New gear, however, is adding virtualization into the switch itself.

Work on intelligent switching was initially spurred on by a group of start-ups, but most of those vendors have been swallowed up by the big boys. The only one left standing is Maxxan Systems Inc. The San Jose-based company's Maxxan MXV500 can be used either as a SAN switch that supports storage applications or as an application platform that supports storage security, virtualization, snapshots and data replication.

McData, too, is getting in on the act. The company has announced some basic intelligence in its Application Services Modules (ASM), which are port-attached to the i10K and are compatible with previous McData 6000 series directors. The McData ASM is a 24-port, 1U (1.75-in.-high) module that attaches to any port of a McData director or switch. It delivers throughput of up to 1 million I/O operations per second. Current modules support virtualization of continuous data protection, VTLs, replication and data mobility. As at AOL, the idea is to accomplish far more without having to rip out cables and rearchitect the SAN.

Dreams Do Come True

According to Christian at Norfolk Southern, however, true SAN virtualization remains a dream. But progress is being made: A team of storage vendors is working with McData and others to add a heterogeneous virtualization layer -- a major missing piece of the intelligent switching puzzle. Within a few months, ASMs attached to an i10K will be able to take advantage of this virtualization layer to perform a wide array of new services.

"I will no longer need one kind of multipathing software for my HDS box and another kind for my EMC array," says Christian. "I'll only need one brand of multipathing software that will be able to function across several different arrays."

While vendors want to hurry the world onto their new intelligent gear, the bottom line is that the Really Intelligent Switch still lies in the future.

"Intelligence has been just over the horizon for quite some time and has yet to make a big splash," says Mike Karp, an analyst at Enterprise Management Associates in Boulder, Colo. "There will also be a delay in market uptake, since these things are serious investments."

AOL's Pollack doesn't plan to get into virtualization and other application intelligence functions until he has gained more confidence on the transportation side.

Norfolk Southern, on the other hand, doesn't even plan to deploy SAN routing until the end of the year. Says Christian, "That gives us another year for the technology to mature."

Robb is a freelance writer in Los Angeles.

SIDEBAR

Intelligence: Where should you put itt?

Depending on which vendor you speak to, you'll hear a different opinion on where storage intelligence and virtualization should really be placed. McData and Brocade, being switch vendors, claim that the switch is the most logical place. HDS, with its focus on disk arrays, insists that intelligence must reside in the storage controller. And Cisco, the king of networking, suggests -- surprise, surprise -- that the network should contain all storage virtualization.

Such a divergence of opinion may well be a factor in the slow corporate adoption of storage intelligence. According to a study of 269 enterprises by Framingham, Mass.-based IDC, only 8 percent are doing any virtualization at all, while 23 percent plan to implement some in the next 12 months. If you focus solely on companies with 10,000 or more employees, however, usage rises to 19 percent, with another 31 percent saying they plan to add a virtual component within a year.

"It's very hard to compare storage virtualization technologies, as they are mostly theories at this point," says Rick Villars, a storage analyst at IDC. "We will need time to see how well they deploy in the real world."