Seung-Ho Choi, CIO at Serome Technology Inc. in Seoul, oversees South Korea's newest and largest voice-over-IP service provider network. With more than 5 million users on the company's unified messaging service, any performance degradation or downtime on its servers and network related to backups equates to lost money. And Serome Technology has nearly a terabyte of e-mail data alone that it must back up regularly.
Choi chose to gamble on new serverless backup technology for his storage-area network (SAN), freeing up his unified messaging servers from having to process the flow of backup information from disk drives to tape. He estimates that his serverless backup software from Legato Systems Inc. in Mountain View, Calif., has cut server-related performance and downtime costs by at least 50 percent.
But Serome Technology is among a tiny minority of enterprises that have chosen to adopt the emerging backup technology.
When SANs began to take hold in large companies more than a year ago, vendors touted serverless backup as the killer application for SANs. However, the technology still has a few kinks that need to be worked out, practitioners say.
SANs greatly increase backup and recovery speeds by moving the backup processes off of the production LAN and onto the faster dedicated Fibre Channel SAN, but backup application servers continue to be a bottleneck as data moves from SAN-attached disks to server memory and across the SAN to tape libraries. A typical backup system can use up to 60 percent of the CPU cycles on a server, according to Meta Group Inc. in Stamford, Conn.
By all estimates, more than 15 percent of Fortune 500 companies are now running data backups on SANs, so serverless backup would seem to be the next logical step.
But the technology is complex. "You can't just throw (serverless backup) in and have it solve a problem. It takes some talent to make it all work together," says Michael Adams, product marketing manager at Veritas Software Corp. in Mountain View, Calif.
And the technology still has limitations. While all systems allow for a full restoration, not all of them allow for selective restoration of individual files or directories. "Most people do restores because someone lost a file or table space in Oracle. So, what's the point in bringing everything back when you only need this small piece?" says Adams.
What's more, serverless restore isn't available, says Meta Group analyst Phil Goodwin. "All available systems today restore the data through the server, not direct from tape to disk, and are therefore not a serverless implementation," he explains.
"We advise our clients to wait on the technology until vendors start to deliver serverless restore capabilities, which we expect to see sometime in 2002," Goodwin adds.
Still, Choi chose not to wait. Traditional server-based backup methods wouldn't be acceptable, "given the amount of data we needed to protect and the service levels we needed to provide," he says.
If the unified messaging service servers lose performance during backups, Choi says, "it will directly impact customer service levels and, ultimately, our bottom line."
The fact that vendors tend to implement serverless backup technology differently also adds to the confusion over serverless backups, as does the fact that backup support for specific applications has been missing.
A serverless backup system generates a backup image, or "snapshot," (or, in some cases, a full copy) of the data. This first requires what vendors call quiescing or suspending running applications and flushing buffers to disk. That requires special agent software, and not all applications are supported. "Database support for these kinds of functions has been slow to reach market," says John Webster, an analyst at Nashua, N.H.-based Illuminata Inc. However, vendors are now beginning to provide such agents.
Once the backup server creates the disk image, it releases the application, creates a disk I/O address block map from the image and sends it to an intelligent "data mover" device, which facilitates the transfer between the disk and tape devices.
Vendors such as Legato use software installed on a special workstation or server to generate a block map of the data and another server to act as the data mover using its own copy process.
Other vendors use the Storage Networking Industry Association's (SNIA) model, which has the backup application server issue a SCSI-3 Extended Copy command to a SAN interconnect device, such as a storage router, to initiate data transfer.
In either case, the backup process isn't entirely server-free. "The backup application on the server still has to catalog the information that needs to be backed up. It sends that in the form of segment descriptors to the intelligent device," says S.W. Worth, technical marketing manager at Austin, Texas-based Crossroads Systems Inc. and a member of the Mountain View, Calif.-based SNIA's Interoperability Task Force.
Since the data to be backed up is stored in blocks, not files, individual files can't be retrieved. Some backup software vendors solve this problem by including a file map that's stored as part of the backup image. However, that function isn't supported in every case. For example, Veritas' NetBackup supports file-level restores for Unix but not for other operating systems, the vendor says.
While serverless backup shows promise, the restore issues are showstoppers for users like Kurt Bahrs, a disaster recovery specialist at Aetna Inc. in Hartford, Conn. Bahrs, who is overseeing a SAN implementation that's expected to be completed by mid-2003, says serverless backup holds the promise of eliminating backup windows that take "tens of hours." But without a complete serverless restore function to complement it, he says, he's not interested.
"We're going to try to automate everything within our SAN environment with all our backups for midrange and NT servers," he says. "The key is having a seamless product to do both (backups and restores)."
The iSCSI Option
Emerging iSCSI-based storage network connections may boost the popularity of serverless backup technology.
Serverless backup has a better chance of being adopted if it doesn't require a complex Fibre Channel SAN, says Illuminata analyst John Webster. The SNIA's proposed Extended Copy standard will enable that by supporting iSCSI networks as they emerge over the next few years.
ISCSI extends SCSI connections across existing IP networks to allow block-level transfers of data between storage devices, Webster explains.
"Serverless (backup) was the application backup scenario that was to be used by everyone to justify installing a SAN. That hasn't quite worked out because of cost and complexity," he says.
"ISCSI will help us get there quicker, in that we extend the reach of the SAN out to servers and workstations that are farther away from the data center. So it makes the whole process more palatable," Webster says.