Method and apparatus for storage device management
A method for provided use in a storage device having a controller, the method comprising: identifying, by the controller, a plurality of logical regions in the storage device; obtaining, by the controller, a respective usage metric for each of the logical regions; updating, by the controller, a translation data structure of the storage device, the translation data structure being updated to map any of the logical regions of the storage device to a respective physical portion of the storage device, the respective physical portion being selected based on the respective usage metric of the logical region, wherein the translation data structure is part of a flash translation layer of the storage device, and the translation data structure is configured to store mapping information between a logical address space of the storage device and a physical address space of the storage device.
1. A method for use in a storage device having a controller, the method comprising:
identifying, by the controller, a plurality of logical regions in the storage device; and
updating, by the controller, a translation data structure of the storage device, the translation data structure being updated by: calculating a respective predicted load of each of the logical regions by executing a machine learning engine, grouping the logical regions into a plurality of groups based on the predicted loads, and mapping each of the group to a corresponding memory block of the storage device, the corresponding memory block being selected based on a wear level of the corresponding memory block and the predicted loads of the logical regions in the group,
wherein the translation data structure is part of a flash translation layer of the storage device,
wherein the translation data structure is configured to store mapping information between a logical address space of the storage device and a physical address space of the storage device, and
wherein the machine learning engine is executed in the flash translation layer of the storage device, the flash translation layer including an interface that is configured to provide logical-to-physical address mapping for the storage device.
2. The method of claim 1 , wherein grouping the logical regions based on the predicted loads causes a distance to be minimized, in any of the groups, between a predicted load of a logical region that has the smallest predicted load in the group and a predicted load of another logical region that has the largest predicted load in the group.
3. The method of claim 1 , wherein each of the groups to a respective memory block of the storage device includes assigning each of the groups to a respective contiguous physical portion of the storage device.
4. The method of claim 1 , wherein each of the logical regions includes a page.
5. The method of claim 1 , wherein the predicted load of any of the plurality of logical regions is based on at least one of (i) a count of write requests that are expected to be received during a future time window, (ii) a count of read requests that are expected to be received during the future time window, (iii) a rate at which write requests are expected to be received during the future time window, (iv) a rate at which read requests are expected to be received during the future time window, and (v) a read-to-write ratio that is expected to occur during the future time window.
6. The method of claim 1 , wherein the machine leering engine is trained based on training data, the training data including a read-to-write ratio of any of the logical regions.
7. A storage device, comprising:
a plurality of memory banks;
a processing circuitry operatively coupled to the memory banks, the processing circuitry being configured to perform the operations of:
identifying a plurality of logical regions in the storage device; and
updating a translation data structure of the storage device, the translation data structure being updated by: calculating a respective predicted load of each of the logical regions by executing a machine learning engine, grouping the logical regions into a plurality of groups based on the predicted loads, and mapping each of the groups to a corresponding memory block of the storage device, the corresponding memory block being selected based on a wear level of the corresponding memory block and the predicted loads of the logical regions in the group,
wherein the translation data structure is part of a flash translation layer of the storage device, and the translation data structure is configured to store mapping information between a logical address space of the storage device and a physical address space of the storage device, and
wherein the machine learning engine is executed in the flash translation layer of the storage device, the flash translation layer including an interface that is configured to provide logical-to-physical address mapping for the storage device.
8. The storage device of claim 7 , wherein grouping the logical regions based on the predicted loads causes a distance to be minimized, in any of the groups, between a predicted load of a logical region that has the smallest predicted load in the group and a predicted load of another logical region that has the largest predicted load in the group.
9. The storage device of claim 7 , wherein mapping each of the groups to a respective memory block of the storage device includes assigning each of the groups to a respective contiguous physical portion of the storage device.
10. The storage device of claim 7 , wherein each of the logical regions includes a page.
11. The storage device of claim 7 , wherein the predicted load of any of the plurality of logical regions is based on at least one of (i) a count of write requests that are expected to be received during a future time window, (ii) a count of read requests that are expected to be received during the future time window, (iii) a rate at which write requests are expected to be received during the future time window, (iv) a rate at which read requests are expected to be received during the future time window.
12. The storage device of claim 7 , wherein the machine learning engine is trained based on training data, the training data including a read-to-write ratio of any of the logical regions.
13. A non-transitory computer-readable medium storing one or more processor-executable instructions, which when executed by a controller of a storage device cause the controller to perform the operations of:
identifying a plurality of logical regions in the storage device; and
updating a translation data structure of the storage device, the translation data structure being updated by: calculating a respective predicted load of each of the logical regions by executing a machine learning engine, grouping the logical regions into a plurality of groups based on the predicted loads, and mapping each of the groups to a corresponding memory block of the storage device, the corresponding memory block being selected based on a wear level of the corresponding memory block and the predicted loads of the logical regions in the group,
wherein the translation data structure is part of a flash translation layer of the storage device, and
wherein the translation data structure is configured to store mapping information between a logical address space of the storage device and a physical address space of the storage device, and
wherein the machine learning engine is executed in the flash translation layer of the storage device, the flash translation layer including an interface that is configured to provide logical-to-physical address mapping for the storage device.
14. The non-transitory computer-readable medium of claim 13 , wherein grouping the logical regions based on the predicted loads causes a distance to be minimized, in any of the groups, between a predicted load of a logical region that has the smallest predicted load in the group and a predicted load of another logical region that has the largest predicted load in the group.
15. The non-transitory computer-readable medium of claim 13 , wherein mapping each of the groups to a respective memory block of the storage device includes assigning each of the groups to a respective contiguous physical portion of the storage device.
16. The non-transitory computer-readable medium of claim 13 , wherein each of the logical regions includes a page.
17. The non-transitory computer-readable medium of claim 13 , wherein the predicted load of any of the plurality of logical regions is based on at least one of (i) a count of write requests that are expected to be received during a future time window, (ii) a count of read requests that are expected to be received during the future time window, (iii) a rate at which write requests are expected to be received during the future time window, (iv) a rate at which read that is expected to occur during the future time window.
18. The non-transitory computer-readable medium of claim 13 , wherein the machine learning engine is trained based on training data, the training data including a read-to-write ratio of any of the logical regions.