The Trouble with Distributed Systems

In a single computer, if something goes wrong, it usually crashes completely. In a distributed system, we have Partial Failures: some parts work, some are broken, and you might not know which is which.

Cloud Computing vs. Supercomputing

• Supercomputing: Checkpoint state frequently. If any node fails, stop the whole cluster and restart from checkpoint. (Reliable hardware).
• Cloud Computing: Service must stay online. Faults are handled in software. (Unreliable commodity hardware).

We must build reliable systems from unreliable components.