"Storage Strategies" by Jon Toigo
A weekly e-mail on storage
Tuesday, November 26, 2002

Is Tape Dead?

A growing number of cheap disk arrays based on IDE/ATA "desktop" drives
are beginning to challenge tape as a backup media. Will tape bite the
big one in this war of byte writes?

The theme keeps emerging in the trade press and in vendor brochures
that tape is dead. Depending on the vintage of the press release one
reads, tape has been relegated to posterity by mirroring, by
virtualization, by SAN, or by the increasing availability of cheap
IDE/ATA disk arrays.

The truth of the matter is somewhat less clear.

Mirroring has not made a dent in the tape market, despite the claims of
mirroring product vendors, and in most cases, continues to coexist with
tape even in those organizations that have applications with critical
data availability requirements. The key reason is that mirroring does
little more than replicate data at high speed. Any data -- even the
corrupt variety.

Generally speaking, mirroring provides one layer of data protection.
Data copies can be made more rapidly to disk targets than to tape
targets, alleviating "backup windows" that make life miserable for
firms with huge datasets to back up. The downside, however, is that
mirroring is expensive -- $1 million or more for a pair of high end
arrays, mirroring software and network interconnects for a 1+ mile
mirror configuration. Mirroring often requires a lock-in to a single
vendor (EMC SRDF will only work between two EMC arrays, IBM XRF will
only work between two or more "Sharks," etc.). Mirroring also
replicates both bad and good data at high speed. Tape is typically
used in addition to mirroring for long-term data storage.

Nor has virtualization killed tape. Virtualization vendors claim that
their methodology for creating, maintaining and managing volumes --
logical representations of physical disk drives -- provides tremendous
flexibility and throughput advantages that can be applied to high speed
data replication applications. In short, writing data to a virtual
disk can provide de facto mirroring in the process if redundant
physical targets are defined to the virtual volume. While this may be
true, the result is still mirroring. Network costs and software costs
(for virtualization, as opposed to mirroring software) persist.
Furthermore, bad data can be replicated in a virtualized environment as
readily as it can within a traditional mirror.

Contrary to some vendor pronouncements, SANs have not killed tape. In
fact, of the 11,000 or so Fibre Channel fabrics deployed to date,
upwards of 85 percent have been deployed explicitly to share a tape
library. Fabrics are seen as an efficient (though extremely pricey and
often unwieldy alternative to) LAN-based data transport: a back-end
"network" option that might make sense for consolidated tape backup
under certain circumstances. In other words, SANs have not displaced
tape backup, but may provide another tape-to-disk interconnect option.

So, then, what about IDE/ATA arrays? Does the increasing availability
of cheap disk obviate the need for tape? Several points need to be

At this writing, tape and disk are still separated by a chasm of media
cost differential. Tape is still a substantially cheaper medium for
data storage than disk. In fact, the reduction of cost differentials
between tape and disk solutions for data protection depend less on
hardware than on software. Both tape and disk media costs increase as
the volume of data to be protected increases, but neither is as
expensive as software costs as data volume increases. Disk-to-disk
replication software is typically sold on a capacity pricing model.
The more data that must be replicated, the more expensive the
replication software and the overall solution -- by a great deal of
dineros in most situations.

That is one reason why early disk-to-disk backup solutions (such as
Quantum's DX30 platform) treat their disk components like tape -- that
is, as another tape backup system target. Such a solution enables
companies to preserve their existing investments in tape backup
software (tape backup software is typically licensed on a per server,
rather than a per GB basis) and related expertise.

For IDE/ATA (or SCSI or Fibre Channel) disk platforms to replace tape,
the platform must be used for more than simple data copy. For example,
Avamar Technologies' recently announced AXION platform provides an
additional service, called "Commonality Factoring" that optimizes the
backup data set, reducing the total size of the data copy. The
solution actually reduces the cost for data replication as the size of
the target data grows.

In the final analysis, just substituting disk for tape to obtain cost
advantages remains a questionable strategy. Look for added value in
the form of software functionality that will help to offset the added
cost of using random access media.

Copyright 2002 101communications LLC. Contact jpowell@101com.com

Questions or problems regarding this web site should be directed to abeckman@outdoorssite.com.

Copyright 2008 Art Beckman. All rights reserved.

Last Modified: March 9, 2008