While the title of this tip may sound schizophrenic, there's some truth to both assertions. On the one hand, using tape for on-site backups can be problematic. On the other, tapes are perfect for sending backups and archives off-site. Someone would have to invent a completely new medium before tape is truly dead.
Is there a better way than tape? Let's first take a brief history of the evolution of tape. The initial development of ultra-fast tape devices also brought with it the need for software that could write to them fast enough. Consequently, multiplexing was born.
Many backup software products use multiplexing (also known as interleaving) to send backup streams from multiple file systems or multiple clients simultaneously to one tape drive. This allows a tape drive to "stream" at its rated speed.
Without multiplexing, many of today's tape drives would never write anywhere near their rated speed. However, restoring data from a multiplexed tape can take much longer.
Storage administrators find themselves in a quandary. They need multiplexing to be able to finish their backups in time, but multiplexing makes restores harder. The only way to turn off multiplexing would be to use a media that doesn't need to stream -- disk.
But disk is too expensive, right? Not anymore. The newest kid on the block is an IDE-based, Fibre Channel, SCSI or FireWire addressable RAID cabinet. Depending on the hardware and software placed in front of it, they can cost only $5,000 to $20,000 per terabyte. (They come in many varieties, from low-end FireWire versions to high-end arrays with NAS heads in front of them.)
Not only is this cheaper than a tape system, some of these arrays are even cheaper than buying replacement media for an existing tape library. Although IDE disks may not be ready for an OLTP database, they could be ideal for on-site backups.
Using a striped file system and disk files as backup devices, such disk arrays could present a backup system with an unlimited number of devices to back up to and none of them would require multiplexing. Once last night's backups are sent to disk, they can then be easily copied to tape and sent off-site. (These tapes would also not need to be multiplexed, because the copy is being made locally.)
While this design may look similar to backup systems that use disk as a cache for tape, it is actually quite different. In traditional backup systems, backups that go to a cache disk are usually deleted off the disk within a few days. In comparison, these backups would be kept on the cache disk as long as they are needed. This disk is so inexpensive that you could actually keep all of your cyclical full and incremental backups on disk. Day-to-day restores would be fulfilled instantaneously by disk. Tape would then only be used for disaster recovery and archival restores.
For environments not currently performing off-site backups, this is a workable alternative. (There are more companies like this than you'd think.) Storage administrators in these environments could start doing on-site backups to one of these arrays, and do their off-site backups to their existing tape library. Now, that's something to think about.
About the author: W. Curtis Preston is the president of the Oceanside, Calif.-based analyst and consulting firm, The Storage Group, Inc. He's made it his business to know about the various backup/recovery tools, techniques and procedures available. Curtis is also presenting the following session, "Uncover Clustering Advancements in Storage Networking," at TechTarget's upcoming free Storage Management conference, March 20-22, 2002, at the Chicago Hilton Hotel.
Additional Resources:
- Do you agree or disagree with the above tip? Maybe you just have a question on a backup/recovery issue. In either case, plan to join Curtis Preston along with our other searchStorage backup/recovery expert, Jim Booth, and a few backup/recovery vendors in our Storage Management Tips & Tricks discussion forum, Wednesday, Feb. 20, 12-5pm EST.
- See also our searchStorage featured topic on Backups: disk vs. tape?
