| 摘要 |
Errors occurring in computing clusters and other computing systems can impact system performance. Each error has an error type and each error type has a base cost estimating importance of correcting the error. Each error type also has a confidence indicating the level of agreement between those who fix the errors and those who assigned the base cost. An error type's actual cost is produced using the base cost and confidence. An error cascade map contains estimates that one error will cause another. An error type that causes other error types has a cascade cost. Upon detecting an error type, a repair order can be generated, depending on the cost involved. Repairs are then performed. Feedback mechanisms and correlations can be used to update the confidences and the error cascade map.
|
