In the backup world today, the conventional wisdom is that disk is fast and tape is slow. While it is usually true that introducing disk into the backup storage chain increases performance, it is not always quite so simple. Architecture and engineering design considerations must still be applied to ensure that the introduction of disk actually meets the desired goals.
"So, how can disk be slower than tape?" you ask. Well, let's start by more accurately characterising the problem with modern tape drives: They are not too slow; they're actually too fast. Drives such as Linear Tape Open-3 can transfer data at 80MB/sec. uncompressed. The problem is that few backup streams, particularly those of higher-overhead, incremental backups, can keep up. Unfortunately, the performance of most tape devices, even those with variable-speed capability, falls off dramatically if data rates can't be sustained, hence the poor backup performance suffered in many environments.
Disk, on the other hand, is far more forgiving. However, it would be a mistake to think that this always results in higher speed. We have encountered situations where well-performing tape libraries were replaced with virtual tape libraries (VTL) but were unable to equal, much less surpass, the performance of the tape that they replaced.
The likely reason for this is a combination of the nature of backup data sets and the architecture of VTLs. Specifically, the backup workload is atypical of the normal usage pattern of disk arrays. It consists of continuous high-volume sequential I/O, whereas most disk arrays are tuned to smaller, more random data mixes of reads and writes. Add to this the fact that VTLs are typically architected in large RAID-5 drive sets to minimise cost. The result can be a reduction in the degree of independent data-stream parallelism of a VTL, when compared with the multidrive tape library it is replacing.
Should people be considering VTLs? Absolutely -- they provide significant advantages. However, before making the investment, spend some time understanding your current environment and the real sources of backup bottlenecks. In particular, watch out for situations where an environment is already able to sustain high, nonmultiplexed tape writes.
Jim Damoulakis is chief technology officer of GlassHouse Technologies Inc., a leading provider of independent storage services. He can be reached at email@example.com.