Apr 14, 2015
Splice Codes (www.splicecodes.com), a new player in Erasure Coding from Norway, tries to promote universal data protection methods for data networking so obviously very aligned with distributed systems needs and challenges. Not easy as erasure coding like replication are considered to be just a feature, certainly important but still a feature. The company developed an interesting approach based on Random Linear Network Coding where the data sources are all encoded and the result packets equal in size the sources meaning that they operate in a non-systematic mode. But they add they can also do systematic where data are in clear are maintained while generating additional data parity packets. According to the literature of Splice Codes, they use a structured approach for creating sparse coefficient matrices based on Latin rectangles. With that, failures during decoding are minimized and packets dependences are maximized helping recovering lost packets. Splice Codes claims to be very efficient in term of computation resource and speed consumption arguing to be one of the best Forward Error Correction technique. They recommend this choice when the network is fast with low-latency. The team recognize that the code has more overhead that Reed-Solomon but with a gain in speed able to decode at 30Gb/s and complexity reduction, they even claim to be faster with even more overhead. The product is available as a C library or for FPGAs and ASICs.
Posted by Philippe NICOLAS à 12:48:00 AM