Understanding Hard Disk Failure Rates

Hard Disk vendors publish a number for some hard disks called mean Time Between Failures (MTBF), and that number has generally risen over the years to be more that a million hours in some cases. User experience, however, seems to differ from expectations, the term can be deceptive, and failure rates can often reach several percent, according to Computerworld on Monday.

One problem is the interpretation of the term MTBF. Itis an average, and that means thereis a 50-50 chance the drive will fail at less than the MTBF. Statistically, many could fail in far less time. As a result, the MTBF term is falling into disrepute, according to Mary Brandel who cited many experts in this area.

"...how do these numbers help a person who wants to evaluate drives?" asked Steve Smith, a former EMC Corp. employee. "I donit think they can."

Stan Zaffos, a Gartner analyst, thinks that while MTBF might be an accurate term in the mind of the vendors, itis hard to translate that into a number thatis meaningful to customers. Robin Harris, an analyst at Data Mobility Group thinks a better number might be Average Failure Rate (AFR), and the industry is using that number in an effort to be less misleading.

"People want to know, in a given year, what percentage of drives they can expect to fail," according to Bianca Schroeder, one of the co-authors of an on-going Carnegie Mellon study.

Even then, there are problems. While vendors typically list a AFR less than 1 percent, the Carnegie Mellon study showed AFRs most commonly 2 to 4 percent. The problem there is how vendors define failed drives.

Carnegie Mellon is working with the vendors to come up with more realistic numbers, but even then there are complicating factors such as workload and environment. One thing that might help is a database of failures of drives of various types. Carnegie Mellon is developing that database with Usenix.

In the end, it comes down to real world user data, and such a database will help customers of all types evaluate their prospective purchase. Until then, however, vendor data sheets and reputation remain foremost.