Posted by
Transferring research results into products and services is always a challenging part of Microsoft Research’s job. No two transfers ever happen quite the same way. This year, Microsoft Research Silicon Valley had a tech-transfer experience that is, I believe, unique in the history of Microsoft Research—at least, I can’t remember one of a similar sort in the 11 years I’ve been here.
The technology in this case is a novel form of erasure codes, called “locally reconstructable codes,” that can make a dramatic improvement in the resources required to provide necessary redundancy in a storage-based service. These codes have substantially better space efficiency than classic Reed-Solomon codes, as well as better performance both in normal operation and during error recovery. It is surprising and gratifying that even in a field that has been explored so extensively, it is still possible to make important discoveries.
