Tiger - Disk-Adaptive Redundancy Without Placement Restrictions

Explore a 16-minute conference talk from OSDI '22 that introduces Tiger, a novel disk-adaptive redundancy system for large-scale cluster storage. Learn how Tiger efficiently eliminates adoption-blocking placement constraints while providing higher space-savings and lower risk compared to previous designs. Discover the concept of eclectic stripes, which allow for tailored redundancy based on diverse failure rates of selected disks. Understand the benefits of Tiger's approach, including compatibility with pre-existing placement policies, improved space-savings, reduced IO burstiness when changing redundancy schemes, and enhanced robustness through increased risk-diversity. Gain insights into the new mean-time-to-data-loss (MTTDL) approximation technique and methods for ensuring safe per-stripe settings in the face of changing device failure rates over time.