This method of encoding supports a single failure. So, if weВ store X, Y, and X + Y, any symbol is recoverable from the other two. The storage overhead is now 1.5x, which is twice as space-efficient as triple mirroring! One solution is to store X + Y as an additional symbol. Suppose weВ want to encode two symbols: X and Y.

This is less space-efficient than encoding multiple data symbols together.ДЄ more efficient method is a Reed Solomon erasure code. In mirroring, each symbol of data is coded independently of the others. But since we store three copies for each symbol of data, the storage overhead is expensive.Д¬an we devise an algorithm with similar availability guarantees but a lower storage overhead? Towards a Smarter Encoding Scheme The probability that all three disks fail on the same day is then about one in a billion (this would be called вЂњnine 9sвЂќ of availability in other contexts). In triple mirroring, we have three copies in total, making this scheme resilient to any two failures.ДЄnd how about fault tolerance? LetвЂ™s assume for simplicity that failures are independent and that a hard drive fails with the probability 1 / 1000 on any given day (corresponding to mean time to failure of about 3 years). Mirroring involves storing identical copies of each piece of data on different disks and is the simplest method of achieving redundancy. The first encoding scheme we used is called triple mirroring. It works by storing redundant pieces of information in a way that allows recovery from complete storage device failures. At Rubrik, the team chose a scheme that achieves a middle ground between the two. Every file system needs to make this tradeoff between availability and overhead. However, the cost of redundancy is increased storage overhead. The more redundancy we add, the greater the fault tolerance. The only way to achieve fault tolerance is through redundancy, which refers toВ storing extra information about the data across different drives to allow for its complete recovery in the eventВ of a failure. In this case, the loss of any single drive should not result in data loss. Suppose weВ want to store a piece of data on a fault-tolerant and distributed file system.

Duplicacy erasure coding