Slivers, Not Replicas: How Walrus Stays Reliable Under Churn
@Walrus 🦭/acc Erasure coding is where Walrus quietly wins its reliability. When a blob lands, it’s transformed into data slivers plus repair symbols, and those pieces are spread across many storage nodes. You’re no longer hoping two or three full replicas survive; you just need enough fragments to reassemble the original. The practical payoff shows up during churn: repairs can target only what’s missing, so recovery is faster and less bandwidth-hungry than copying whole files.