Thursday, September 18, 2008

Distributed System Models

Fail-stop
Process abstraction: crash-stop process.
Link abstraction: perfect link.
Failure detection: perfect failure detector.

Fail-silent
Process abstraction: crash-stop process.
Link abstraction: perfect link.
Failure detection: no failure detector.

Fail-noisy
Process abstraction: crash-stop process.
Link abstraction: perfect link.
Failure detection: eventually failure detector.

2 comments:

امیرحسین said...

Do we really need such a fine-grained classification ? Well, I hadn't seen such definitions before, (I usually see fail-sop and crash-stop being used). In particular, I don't understand the benefit of defining Fail-silent.
Anyways, good job!

Francesco Bongiovanni said...

Indeed we do need different models. As for the fail-silent model, it supposes a crash-stop model for the processes, but does not assume any failure detection. So in few words, when a process crashes, other alive processes have no means to get any information about the crashed process.