In an asynchronous system where one process may crash, which two properties can no deterministic algorithm both guarantee?
- Encryption and authentication
- Low latency and high throughput
- Agreement and finishing
What model does the FLP impossibility result for consensus assume?
- Synchronous rounds with no failures
- Async messages with one allowed crash
- A single computer with no network at all
With no bound on message delay, a slow process and a crashed process can always be told apart.
Circle one: True False
What is a timeout standing in for?
- A faster network cable
- A failure detector
- A bigger hard drive
A node waits for a reply that never comes. What is the dilemma?
- Waiting may never finish; deciding may disagree
- Its silence proves a crash, so it must stop the cluster
- Its own clock is wrong, so it must reset before waiting
A timeout fires on a node that was merely slow. What went wrong?
- The timeout proved the node crashed, so nothing went wrong
- The node broke the timeout
- The detector suspected wrongly, which timeouts can do
Your system must keep agreeing despite slow networks. Which concession fits?
- Give up guaranteed termination and use timeouts
- Give up agreement whenever waiting hurts
- Ban all timeouts as unreliable
A vendor's tests never saw its deterministic consensus algorithm fail, so it promises guaranteed agreement and guaranteed finish under asynchrony with one possible crash. How do you answer?
- Reject: no deterministic algorithm can promise both
- Accept, since enough testing can show the promise holds
- Accept, if the tests ran on more than three machines