Magazine Article | June 1, 2003

Who Says You Can't Afford A SAN - Or Two?

Source: Field Technologies Magazine

With the iSCSI (Internet SCSI) specification now secure, you can confidently build IP-based storage networks. Never had a SAN in the plan? Think again.

Integrated Solutions, June 2003

If your organization is a small to midsized enterprise, chances are your IT resources aren't unlimited. No company's are. Unfortunately, your rapidly accumulating corporate data is raising red flags in terms of your storage infrastructure's flexibility and scalability. Given all of the direct attached storage hanging off your servers, you can't access half of the capacity that's sitting out there. Not only that, but your backup/restore operations are taking way too long and are threatening uptime. And, disaster recovery? Yeah, right.

To tackle these problems, you'll need to reconfigure your storage as a shared, networked environment. A SAN (storage area network) is the obvious answer. But, the performance benefits of Fibre Channel-based SAN solutions are overkill for your operations, you have decided, especially considering the cost. And, even if you could afford to build Fibre Channel SANs, the resulting management woes would likely crush your understaffed IT department. And, anyway, no one on that team really has any experience or expertise in deploying and maintaining Fibre Channel SANs. What you need is an alternative.

For some time now, IP (Internet Protocol)-based storage has been touted as the way to bring affordable, manageable storage networks to the SME (small to medium enterprise) market. While various connectivity protocols, including FCIP (Fibre Channel over IP), have been proposed as the ideal transport engine, the most likely candidate for widespread adoption remains iSCSI (Internet SCSI [small computer system interface]). iSCSI doesn't require Fibre Channel networking (though it can supplement it). It takes advantage of a company's existing Ethernet LAN or WAN (wide area network) infrastructure and uses SCSI interfaces between servers and storage. Ethernet and SCSI are, of course, familiar, solid technologies well understood by most, if not all, IT professionals. Chalk up one - no, two - for the home team.

Now that the iSCSI standards development has been completed by the Internet Engineering Task Force (IETF), more and more iSCSI initiator and target devices will be coming to the market. With the ratification of the iSCSI specification driving product development, you'll be able to confidently move forward with iSCSI-based storage networking initiatives.

Open Windows For iSCSI
The ratification of the standard is key to market adoption, first and foremost because of one key player - Microsoft. The software vendor recently announced it will support the ratified iSCSI specification. That endorsement is critical. It means iSCSI can be seamlessly integrated with devices operating in Windows-based server and storage environments, which are common in the SME space.

Industry insiders agree that the Microsoft announcement is crucial. Says John Matze, president and CEO of iSCSI vendor Okapi Software, Inc. (Poway, CA), "The ratification of the spec matters for just one reason: Microsoft will now have iSCSI drivers. There are a lot of Microsoft shops that couldn't afford to go to Fibre Channel but would like to have the high-end functionality." According to Glenn Clowney, director of strategic marketing for iSCSI vendor Adaptec Inc. (Milpitas, CA), "For us, the Microsoft announcement allows us to put a check mark in one big box on the list of factors that will drive adoption. It will ease the minds of a lot of customers." Bill Huber, CTO at iSCSI vendor StoneFly Networks (San Diego), points to Microsoft's increasing integration of storage functionality in its server management features. "They've defined an API [application program interface] for creating mirrors and growing storage," says Huber. "When people start to understand Microsoft's new services, there's going to be a real jump in the realization of how well iSCSI fits the midmarket. For those midsized companies, there will now be services that were previously seen only in big box solutions, such as Symmetrix."

Alleviate The I/O Crunch
In addition to using existing resources, there are benefits to routing iSCSI-based storage traffic over existing (or upgraded) Ethernet LANs and WANs (wide area networks). First, iSCSI enhances the ability of TCP/IP (an open computer communications language)-based processing to extend the distance SCSI commands can be carried. Typically, a SCSI bus can connect servers to storage at distances of only a few yards away. With TCP/IP, those distance limitations are eliminated, enabling centralized storage networks and subnetworks, as well as remote disaster recovery environments. (Of course, for transmissions between local servers and a remote location, security issues enter the picture.)

However, TCP/IP loads often require heavy-duty, high I/O (input/output) activity on the servers. And, that's where iSCSI steps in again - to offload TCP/IP processing from the server's CPU (central processing unit). That allows processing power to be directed at what it was primarily purchased to do - run applications and databases. "TCP/IP is a major hog as far as CPU utilization," says Clowney. "If 60% or 70% of a server's cycles are being used just to move data around, you're not getting optimal application performance. But, if you use iSCSI to offload TCP/IP processing from the general purpose CPU, you can free up 80% or 90% of your server's cycles for application processing."

The kind of offloading Clowney describes is particularly useful for lessening the burden on transaction-oriented servers running, for instance, high-end Oracle or SQL databases. For those applications, a TCP/IP offload engine (TOE) is necessary. But, according to Matze, few lower end (in terms of I/O) applications are likely to require a TOE. That's because unlike, say, the CIFS (common Internet file structure) protocol running on a NAS (network attached storage) box, iSCSI can make full use of disk caching. "A NAS box is a sharable device. Anytime you want to access a subdirectory, you have to see if anything has changed," Matze explains. "But, in the iSCSI world, the server typically thinks the storage is local, so it will cache much of the directory structure and a lot of the most frequently used files. So, it has to go out to disk fairly infrequently."

Back Up Here, Back Up There
Proponents of iSCSI often describe grand, visionary uses, such as remote mirroring to a full-blown dual SAN for disaster recovery (which the protocol can indeed support). However, many companies will immediately use iSCSI for a more mundane yet still fundamental task: backup/restore. iSCSI is a key enabler for disk-based backup, a growing trend. "In a midsized organization, you may have a small, 8-cartridge tape autoloader sitting behind a server, with backup taking most of the night," says Huber. "An alternative is to have a storage pool and do disk-to-disk [D2D] backup. You can back up, say, five servers in an hour, which frees up the servers and gives you 23 hours to back up to tape."

Matze agrees and notes that iSCSI's ability to eliminate SCSI's distance limitations will be a boon to both disk-based and remote backup initiatives. "Tape needs to be there for long-term archiving," Matze says. "But, people will want to use disk to maintain their critical restore time frames. And, iSCSI allows you to move data from point to point very quickly and efficiently - even remotely."