The company has developed a unique data protection service named DEEPR for Distributed Erasure Encoding with Priority Repair. Without detailing known drawbacks about RAID limitations, rebuild time, large drives, distributed topologies… let’s understand what Igneous has built.
This is a key point to differentiate Igneous with some other solutions that try to clone and mimic that philosophy without any patent in erasure coding but just implement a classic basic Reed-Solomon.
A few elements are really important when you design such erasure code (EC) model. First, of course you project a hardware overhead ratio and it has to be good – between 1.2 and 1.8 – to be compared against 3x mirroring, this is the financial argument. Second the number of failures you can support especially in parallel or concurrently, this is the reliability and availability advantages and finally what is the I/O behavior when you write/read data, this is the transparency and application QoS effect. This is why some players manage the priority of rebuilding process to reduce the impact on production but the rebuild time lasts longer.
For Igneous, each data server named dataBox shipped and deployed has 60 drives in it organized with 2 groups of 28 drives for 212TB of usable storage space. Each of these 28 are organized as 5 local groups of 5 disks with local EC 4+1 (5 local parities) with 3 additional parities named global. Immediately it means that you can loose 1 to 8 concurrent drives in a set of 28. Wow, 1.4 (=28/20) is a pretty efficient hardware overhead ratio within one dataBox. When you lose a disk, you just need to read 4 drives aka localized repair and not a large data set. And each group can support a loss of drive without impacting a global set. When you have multiple failures within a group you need to consider the global parities. What is really new for Igneous is covered by Patent #9116833 with global erasure codes. This makes great differences with other cloud storage solution supported by real invention and patents.