Machine learning error prediction in storage arrays
AssigneeSeagate Technology LLC
InventorSunil Savanur
Life of this patent
Estimating the term…
Abstract
Systems and methods for machine learning error prediction in storage arrays are described. In one embodiment, the method includes monitoring events of at least a first storage drive array, analyzing the monitored events using machine learning, identifying failure events based at least in part on the analysis and operating a prediction model engine to predict potential errors in storage drive arrays, the prediction model engine being built based at least in part on the analysis of the monitored events or the identifying of the failure events, or both.
Claims19
A system comprising:
- one or more processors;
- memory in electronic communication with the one or more processors, the memory storing computer executable instructions that when executed by the one or more processors cause the one or more processors to:
- accumulate monitored events into the memory, the monitored events associated with operation of a first storage drive array to transfer data between storage drives in the first storage array and a host device;
- arrange a first portion of the monitored events into a training set and a second portion of the monitored events into a test set;
- analyze the training set of monitored events using machine learning to generate an analyzed training set that identifies at least one failure pattern in the monitored events;
- operate a prediction model engine to predict potential errors in a second storage drive array responsive to an analysis, by the prediction model engine, of the test set of monitored events and the analyzed training set, to feed at least a portion of an output of the prediction model engine back into an input of the training set to provide an updated training set, and to analyze the updated training set using the machine learning.
A method for error prediction in storage systems, the method comprising:
- monitoring events of a first storage drive array during operation of the first storage drive array to transfer user data between storage devices of the first storage drive array and a host device;
- accumulating the monitored events into a computer memory;
- dividing the accumulated monitored events into a respective training set and test set in the computer memory;
- forwarding the training set for analysis by a machine learning circuit to generate an analyzed training set that identifies at least one failure pattern;
- transmitting the analyzed training set and the test set to a prediction model engine circuit to generate a predictive model of behavior of the first storage drive array, the prediction model engine circuit operative to use a portion of an output thereof as an additional input to the training set to provide an updated training set;
- and
- using the predictive model of behavior from the prediction model engine to predict a potential failure condition for a different, second drive array based on monitored events obtained from the second drive array during operation to transfer user data between storage devices in the second drive array and a host device.
A computer-program product for error prediction in storage systems, the computer-program product comprising a non-transitory computer-readable medium storing instructions thereon, the instructions being executable by one or more processors to perform the steps of:
- monitoring events of a first storage drive array responsive to transfers of user data between storage drives in the first storage drive array and a host device;
- analyzing the monitored events by subjecting a first portion of the monitored events to a machine learning algorithm to detect at least a first performance pattern of the storage drives, by subjecting a different second portion of the monitored events and the detected at least one performance pattern to a prediction model engine to generate an output, and by feeding a portion of the output back to an input of the training set to provide an updated training set to detect at least a second performance pattern of the storage drives;
- and
- using the prediction model engine to predict a potential failure condition in a second storage array responsive to monitored events obtained during user data transfers between second storage drives in the second storage array and a host device.
Dependencies are inferred from the claim text.