Disk replacement using a predictive statistical model
AssigneeAmazon Technologies, Inc.
InventorsRoss Bevan Engers, Stefan Letz
Life of this patent
Estimating the term…
Abstract
In a provider network, attributes of one of a plurality of storage devices of the provider network are identified for failure monitoring. Based on a failure prediction model, a predicted probability of failure of the selected storage device is determined. The failure prediction model is based on historical and current data associated with failures of the storage devices of the provider network that have common attributes. The selected storage device is deactivated in response to determining that the predicted probability of failure of the selected storage device meets a criterion.
Claims23
A computer-implemented method for managing storage devices of a provider network, the method comprising:
- selecting one or more storage devices for which a predicted probability of failure is to be determined;
- identifying common attributes of the selected storage devices;
- accessing historical and current data associated with failure of the selected storage devices, the historical and real-time data selected based at least in part on the common attributes;
- calculating the predicted probability of failure based at least in part on the historical and real-time data associated with the failure of the selected storage devices and a failure prediction model;
- deactivating the selected storage devices in response to determining that the selected storage devices have a predicted probability of failure that meets one or more criteria, the one or more criteria comprising a probability of failure during a period of time indicated by the criteria;
- and
- updating the failure prediction model as updated data associated with the failure of the storage devices becomes available.
A system configured to manage storage devices in a provider network, the system comprising:
- at least one memory having stored therein computer instructions that, upon execution by one or more processors of the system, at least cause the system to:
- identify attributes of one of a plurality of storage devices of the provider network for failure monitoring;
- determine, based at least in part on a failure prediction model, a predicted probability of failure of the one of the plurality of storage devices, wherein the failure prediction model is based at least in part on historical and current data associated with failures of the plurality of storage devices that have common attributes;
- and
- identify for replacement the one of the plurality of storage device in response to determining that the predicted probability of failure of the one of the plurality of storage devices meets a criterion, the predicted probability of failure being indicative of a likelihood of failure of the one of the plurality of storage devices during a threshold period of time.
A non-transitory computer-readable storage medium having stored thereon computer-readable instructions, the computer-readable instructions comprising instructions that upon execution on one or more computing devices, at least cause the one or more computing devices to:
- based at least in part on a failure prediction model, determine a predicted probability of failure of a storage device that is allocated to a customer of a provider network, wherein the failure prediction model is based at least in part on past failure data associated with storage devices of the provider network that have one or more common attributes;
- and
- allocate a different storage device to the customer in response to determining that the predicted probability of failure of the allocated storage device meets at least one service level criterion associated with the customer, the at least one service level criterion associated with the predicted probability of failure occurring within a threshold period of time.
Dependencies are inferred from the claim text.