What Does Russian Roulette and RAID-5 Have in Common?
Answer: if you play too long, you lose.
I found myself talking to a customer this week about whether they should upgrade their backup appliance. I was talking about all of the sexy new features, like data deduplication, that were available on the latest generation backup appliance family. But I then learned that the customer had a large appliance (more than 4 drives) that was based on RAID-5 drives. That pretty quickly changed the nature and the tone of the conversation.
There have been a slew of articles concerning the problems of RAID-5 and increasing drive sizes. The most famous of these was a ZDNet article by Robin Harris – it spawned a tremendous amount of attention – if you google on terms in the article you can find a lot of other articles referencing that. And while technical people argued at the edges of the issues the article raised, the simple facts were that the mathematics were compelling. It boils down to an issue with the number of bits on average that may be read from a given capacity disk drive before an unrecoverable read error occurs.
This was the fundamental reason that Unitrends went to RAID-6 on its higher end systems in 2009 and completely eliminated RAID-5 systems for our 2U and 3U form factors earlier this year. Across our large population of customers, what we found was that while RAID-5 was less expensive (because it only requires one parity drive instead of two), that the problem was that after one drive failed we were seeing second drive failures during the longer rebuilds necessary with larger drives.
If you’re interested in where RAID-6 starts having the same issues as RAID-5 given disk drive growth rates, there was a follow-on article by the same author earlier this year entitled Why RAID 6 Stops Working in 2019. I liked the article – 9 years are several lifetimes in terms of technology – and it validates that for the next few years RAID-6 will remain the preferred technology for larger storage systems.
In any case, if you’re using RAID-5 today in a system with more than 4 total drives, you need to be planning to move to RAID-6. Technologies that reduce redundant data, such as data deduplication, make this even more important.
Or to put this as bluntly as possible, don’t play Russian Roulette with your data with a six-cartridge revolver in which five bullets have been loaded.
