Paper Title: Challenges of Distributed Erasure Coding

Paper Abstract: Excelero's implementation of distributed block device relies solely on client side algorithms without central synchronization mechanism. This allows extremely fast data path without software bottlenecks. However, an immediate implication is that when a few clients run I/O to a shared block device each of them may see the device's topology differently (Degraded mode/Rebuilding/OK/etc). This trait, when combined with redundancy algorithms (RAID-6 erasure coding) introduces various interesting mathematical models, and challenges. A few examples are distributed definition of a "write hole," a completely different meaning of journaling, rolling forward failed transactions, and much more. In this lecture, I will describe the distributed algorithmic model of Excelero block device, the challenges and solutions related to protection RAID redundancy.

Paper Author: Daniel Herman Shmulyan, Data Service Team Leader, Excelero

Author Bio: I love challenges. Can't breathe or live without them. New complicated systems, original solutions, innovations, improvement of mechanisms and more... I am usually involved in computer vision projects but other tools like hammer, volt-meter, negotiation, brainstorming are also useful Specialties: * Computer Vision: Object Detection & Recognition, Face technology, OCR. * Entrepreneurship, Original out of the box thinking, * System Engineering, Algorithmic research, Applied mathematics, Machine Learning