A Failure Detector That Gives Information on the Degree of Confidence in the System
Résumé
This work proposes a new and flexible unreliable
failure detector, denoted Impact Failure Detector (FD), whose
output gives the trust level of a set of processes. By expressing
the relevance of each node by an impact factor value as well as
an acceptable margin of failure in the system, the Impact FD
enables the user to tune the failure detection configuration in
accordance with the requirements of the application: in some
scenarios, the failure of low impact or redundant nodes does not
jeopardize the confidence in the system, while the crash resulting
from a high impact factor may seriously affect it. Either a softer
or stricter monitoring is thus possible. Performance evaluation
results using real PlanetLab [1] traces confirm the degree of
flexibility of our failure detector and that, due to the margin of
failure, the number of false responses may be reduced when it is
compared with traditional unreliable failure detectors.