A method of predicting failure of an information handling device, such as a server, by monitoring an error rate, i.e., n errors per error period. Errors are reported only if the error rate is exceeded. An error count is kept, and errors are leaked from the count if the time difference between errors is more than the error period.