Benchmarking with Churn

Jump to bottom

Jocelyn Thode edited this page Oct 31, 2016 · 4 revisions

Synthetic Churn

We should run 3 scenarios:

Kill one peer per minute (Don't kill the coordinator)
- Use docker kill
Kill one peer per minute (Kill the sequencer for JGroups)
- How should we find the coordinator easily ? (First member of JGroups view?) => We put it in an other service
Kill one peer + Add one peer every 2 minutes
- A problem to fix, how to add a peer (with a new IP preferably) (use SIGSTOP)
- Docker service update randomly kills other services

Problems

When killing the coordinator the first time we can't keep track of who the new coordinator is (it might get killed again) is it a big deal ?

Real Churn

Take a look at FTA