Erasure coding, in practice
Why 10+4 usually beats 3× replication, and the three places where it quietly stops being worth it.
Short write-ups on distributed systems, object storage internals, and whatever else I happen to be taking apart this month.
Why 10+4 usually beats 3× replication, and the three places where it quietly stops being worth it.
A short tour of write barriers, drive caches, and the failure modes that survive a "durable" write.
Two incidents caused by a CDN cache key that looked correct in every dashboard we had.
What I actually look at first when a p99 graph moves and nothing obvious changed.