Predictive data orchestration in multi-tier memory systems
A computing system having memory components of different tiers. The computing system further includes a controller, operatively coupled between a processing device and the memory components, to: receive from the processing device first data access requests that cause first data movements across the tiers in the memory components; service the first data access requests after the first data movements; predict, by applying data usage information received from the processing device in a prediction model trained via machine learning, second data movements across the tiers in the memory components; and perform the second data movements before receiving second data access requests, where the second data movements reduce third data movements across the tiers caused by the second data access requests.
1. A device, comprising:
a first memory component having a first tier of performance;
a second memory component having a second tier of performance; and
a logic circuit coupled to the first memory component and the second memory component and configured to:
receive first data identifying first data movements, between the first memory component and the second memory component, in response to a processor accessing the first data;
provide the first data as input to a prediction model, trained via machine learning, to generate a result as output from the prediction model;
identify, based on the result, second data to be accessed by the processor and second data movements, between the first memory component and the second memory component, to be caused by the second data being accessed by the processor; and
initiate, before the processor accessing the second data, the second data movements.
2. The device of claim 1 , wherein the first data includes data usage information.
3. The device of claim 2 , wherein the prediction model includes an artificial neural network trained via machine learning using the data usage information and a hybrid reinforcement learning technique.
4. The device of claim 3 , wherein the result is generated in real time before the processor accessing the second data.
5. The device of claim 1 , wherein the logic circuit is further configured to:
perform third data movements in response to the processor accessing the second data; and
train the prediction model based on the third data movements.
6. The device of claim 1 , wherein the logic circuit is further configured to:
group data blocks that are used together to identify a mapping between data blocks and data objects, wherein the result is based at least in part on the mapping.
7. The device of claim 1 , wherein the logic circuit is further configured to:
adjust the prediction model based on a performance measurement of the first memory component and the second memory component in servicing data access requests from the processor.
8. The device of claim 7 , wherein the performance measurement is a cache hit ratio of data access requests.
9. The device of claim 1 , wherein the device is configured to be connected to a controller via a bus; and the first data movements and the second data movements are not through the bus.
10. A method, comprising:
operating a first memory component of a device at a first tier of performance;
operating a second memory component of the device at a second tier of performance;
receiving first data identifying first data movements, between the first memory component and the second memory component, in response to a processor accessing the first data;
providing the first data as input to a prediction model, trained via machine learning, to generate a result as output from the prediction model;
identifying, based on the result, second data to be accessed by the processor and second data movements, between the first memory component and the second memory component, to be caused by the second data being accessed by the processor; and
initiating, before the processor accessing the second data, the second data movements.
11. The method of claim 10 , wherein the first data includes data usage information; and the prediction model includes an artificial neural network trained via machine learning using the data usage information and a hybrid reinforcement learning technique.
12. The method of claim 10 , further comprising:
performing third data movements in response to the processor accessing the second data; and
training the prediction model based on the third data movements.
13. The method of claim 10 , further comprising:
grouping data blocks that are used together to identify a mapping between data blocks and data objects, wherein the result is based at least in part on the mapping.
14. The method of claim 10 , further comprising:
adjust the prediction model based on a performance measurement of the first memory component and the second memory component in servicing data access requests from the processor;
wherein the performance measurement is a cache hit ratio of data access requests.
15. A system, comprising:
a processor;
a controller;
a first memory component having a first tier of performance;
a second memory component having a second tier of performance; and
a logic circuit coupled to the first memory component and the second memory component and configured to:
receive first data identifying first data movements, between the first memory component and the second memory component, in response to the processor accessing the first data;
provide the first data as input to a prediction model, trained via machine learning, to generate a result as output from the prediction model;
identify, based on the result, second data to be accessed by the processor and second data movements, between the first memory component and the second memory component, to be caused by the second data being accessed by the processor; and
initiate, before the processor accessing the second data, the second data movements.
16. The system of claim 15 , wherein the first data includes data usage information; and the prediction model includes an artificial neural network trained via machine learning using the data usage information and a hybrid reinforcement learning technique.
17. The system of claim 15 , wherein the logic circuit is further configured to:
perform third data movements in response to the processor accessing the second data; and
train the prediction model based on the third data movements.
18. The system of claim 15 , wherein the logic circuit is further configured to:
group data blocks that are used together to identify a mapping between data blocks and data objects, wherein the result is based at least in part on the mapping.
19. The system of claim 15 , further configured to:
adjust the prediction model based on a performance measurement of the first memory component and the second memory component in servicing data access requests from the processor, wherein the performance measurement is a cache hit ratio of data access requests.
20. The system of claim 15 , further comprising:
a bus configured to connect the controller to the first memory component and the second memory component, wherein the first data movements and the second data movements are not through the bus.