We developed a hybrid probabilistic and possibilistic technique for assessing the risk of an SLA for a computing task in a cluster/grid environment. The probability of success with the hybrid model is estimated higher than in the probabilistic model since the hybrid model takes into consideration the possibility distribution for the maximal number of failures derived from a resource provider’s observations. The hybrid model showed that we can increase or decrease the granularity of the model as needed; we can
the estimate of the P(S*=1) by making a rougher, more conservative, estimate of the more unlikely events of (M+1, N) node failures. We noted that M is an estimate which is dependent on the history of the nodes being used and can be calibrated to “a few” or to “many” nodes.