Natural disasters and government regulations have put pressure on companies to back up data farther away from their main data centers than they do now, but achieving that goal can be pricey. Tape is one option, but it's slow to restore. Dedicated networks can replicate data over a WAN, but at a bandwidth price.
In their quest for cheaper alternatives, some companies are turning to a still-developing option. Storage over IP, or the replication of block-level data over leased virtual private networks, allows users to select the type of wide-area service that best meets their budget and application requirements.
"It's a lot faster than tape. And IP is a lot less expensive than a dedicated line," says James Opfer, an analyst at Gartner.
The three main long-distance transmission protocols -- Internet SCSI (iSCSI), Fibre Channel over IP (FCIP) and Internet Fibre Channel Protocol -- are all significantly different, but they provide a common function: transporting block-level storage over an IP network, which allows administrators to use inexpensive Ethernet-TCP/IP networks to transport and access storage over LANs, WANs or metropolitan-area networks.
ISCSI, which transmits block-level data directly over Ethernet, is primarily a server-to-storage interconnect within the data center. It takes advantage of ubiquitous Ethernet networks to consolidate Windows and Linux server backups by connecting LANs to storage-area networks (SAN). New vendors are now offering arrays that use iSCSI as a tunneling protocol to transmit over long distances.
The greatest savings in using IP to transport data between data centers comes from reduced bandwidth. Long-distance replication of block-level data can use IP, Synchronous Optical Network (Sonet) and Asynchronous Transfer Mode (ATM) networks to transmit data, with little difference in price as long as it's not a dedicated line.
"It's not the hardware. Every time I talk to people, they say it's the cost of bandwidth services," says Opfer.
Storage over IP provides a cheaper, faster alternative for data backup.
But exact savings are hard to pin down: Opfer says bandwidth costs can vary widely by region. If you're using "dark fiber," or surplus fiber-optic cable, "you're going to get wild swings in the price," he says. And with IP, there's no need to hire Fibre Channel network administrators or buy switches or host bus adapters to create the network interface.
But financial firms and other companies dealing with sensitive data are still squeamish about adopting a storage-over-IP framework, says Dianne McAdam, an analyst at Data Mobility Group LLC in Nashua, N.H. "What I do see changing is the small to medium-size businesses, where in the past they looked at doing replication and then got the price tag and said, 'Oh my God,' " she says.
Some large companies say the cost benefits outweigh the risks. For example, Don Bolton, manager of infrastructure and systems services at Teppco Partners LP, an energy company in Houston, understands costs issues associated with disaster recovery. Teppco had been paying US$250,000 to US$500,000 a year to SunGard Data Systems, to ensure that should the company's systems go down, they could power up in another region without losing data.
Earlier this year, Bolton says, he decided that Teppco could perform that chore itself by replicating mission-critical, block-level data via a T1 line between data centers in Houston and Tulsa -- a distance of about 500 miles. That move has cut his costs by more than half with a quick return on investment. "I was able to take an existing facility we had and put US$200,000 in it and install like servers with a 16-month payback on our investment. And it's allowed us to grow our infrastructure," Bolton says.
Teppco asynchronously replicates block-level data from Oracle and SQL databases that reside on clustered arrays from Network Appliance that use NetApp's SnapMirror application. "It's pretty interesting technology. If I was a superlarge shop at Continental Airlines or something, with hundreds of terabytes, I don't know if this would be the solution. But with this size, it is perfect," Bolton says.
In asynchronous mode, SnapMirror provides read-only, online replication of a source file system (volume or quota tree level). In the event of a disaster, Bolton says the mirror site could be used as the primary, converting the SnapMirror replica into a writable file system and replacing the original.
Edward Jones & Co has begun to reap the rewards of three years of work and an $80 million investment in a storage-over-IP network [QuickLink 49629]. The system is replicating up to 700GB of data daily to a fully redundant backup site located 1,600 miles away from the brokerage's main data center, near St. Louis. Edward Jones has created a secondary site in Tempe, Ariz., that's a replica of the primary data center, complete with 450TB of SAN-based data.
Bill Hayden, director of data services at Edward Jones, which is using FCIP over ATM to tunnel its data across the U.S., says prices have continued to drop on network bandwidth. "It doesn't matter whether it's running IP, Sonet or ATM," the prices are the same, he says.
The payback? All of Edward Jones' Oracle databases are replicated nationwide in under five seconds. File services are replicated every four minutes, and mainframe databases are typically replicated in 30 minutes. The company says it can flip a switch and begin using its secondary site as its primary one.
"Our sticking point is our [VSAM flat files] on our mainframe," Hayden says. VSAM flat files still take four hours to replicate, a problem that Hayden says could be resolved by updating Edward Jones' EMC Symmetrix arrays and replication software.
"There are always issues," he says. "We're always looking to get the data there faster and looking for ways to get the data to the point where we can update it at either site."