Lessons from the Backblaze Data
Thirteen years of daily records from half a million disk drives: what they say about failure, what they cannot say, and what a storage simulator should assume.
Notes & essays
Notes on computing, research, academic history, and life.
Thirteen years of daily records from half a million disk drives: what they say about failure, what they cannot say, and what a storage simulator should assume.
Faith, mathematical proof, and the engineering cost of algorithms that promise their advantages beyond the reach of any real computer.
I proposed an automata theory textbook to John Carroll when I was an undergraduate. The book now has a restored edition and a new solutions manual.
Witold Litwin’s pioneering work on linear hashing, multidatabase systems, scalable distributed data structures, and many other fields—and the friendship that grew from it.
Encoding a file as its differences from an earlier version, in linear time, and rebuilding it in the space the old version occupies.
The mathematical limits of LLM guardrails, the risks of letting models act, and the people who must answer when safeguards fail.