Adaptive device quality of service by host memory buffer range
A system and method for reducing performance penalties of a host that is supplying a host memory buffer (HMB) for use by a storage device. The method may include modeling desired HMB access timing by the storage device in an initial offline analysis for multiple classes of workloads, periodically updating the access timing data stored in the storage device based on actual use and using the current HMB access timing information to modify storage device access to the HMB on the host. The system may include a storage device controller that quantifies different HMB access timing for different host workloads based on individual HMB regions.
1. A non-volatile memory system comprising:
an interface configured to place the non-volatile memory system in communication with a host comprising a host memory buffer and a command queue;
a non-volatile memory; and
a controller in communication with the non-volatile memory and the interface, wherein the controller is configured to:
determine a first delay period and a second delay period associated with a workload class of a set of commands in the command queue on the host;
add the first delay period to all accesses made by the controller to a first memory region of the host memory buffer to execute the set of commands;
add the second delay period to all accesses made by the controller to a second memory region of the host memory buffer to execute the set of commands; and
subsequent to executing the set of commands, dynamically modify one or both of the first delay period and second delay period to account for a variation between a time spent executing the set of commands and an expected time to execute the set of commands;
wherein:
the first delay period differs from the second delay period; and
the first memory region differs from the second memory region.
2. The non-volatile memory system of claim 1 , wherein:
the host memory buffer comprises a volatile memory; and
the first memory region contains a first type of data.
3. The non-volatile memory system of claim 2 , wherein:
the first type of data comprises logical-to-physical mapping table data; and
the second memory region comprises firmware overlay data for the non-volatile memory system.
4. The non-volatile memory system of claim 1 , wherein the controller is further configured to dynamically change the first delay period in response to receipt, from the host, of latency measurements for commands sent to the host memory buffer from the controller.
5. The non-volatile memory system of claim 1 , wherein the controller is further configured to periodically adjust delay periods for controller access to each of a plurality of memory regions of the host memory buffer in response to latency measurements received from the host.
6. The non-volatile memory system of claim 1 , wherein the set of commands comprises commands to read, write, or trim data in the non-volatile memory.
7. The non-volatile memory system of claim 6 , wherein a command in the set of commands comprises at least one operation including an access to the first or the second memory region of the host memory buffer by the controller.
8. The non-volatile memory system of claim 1 , wherein the controller is further configured to, subsequent to executing the set of commands:
compare the time spent executing the set of commands with the expected time to execute the set of commands.
9. The non-volatile memory system of claim 1 , wherein the non-volatile memory comprises a substrate formed with a three-dimensional memory structure.
10. A method of processing memory accesses, the method comprising:
performing the following in a non-volatile memory system comprising an interface configured to place the non-volatile memory system in communication with a host comprising a host memory buffer and a command queue:
determining, in an offline modeling process, a first delay period and a second delay period;
identifying a workload class fora set of commands in the command queue on the host to be executed by a controller of the non-volatile memory system, wherein the workload class is associated with the first delay period and the second delay period;
adding the first delay period to accesses made by the controller to a first memory region of the host memory buffer to execute the set of commands;
adding the second delay period to accesses made by the controller to a second memory region of the host memory buffer to execute the set of commands;
receiving host response latency data from the host in response to a variation between a time spent executing the set of commands and an expected time to execute the set of commands; and
in response to receiving the host response latency data, modifying one or both of the first delay period and second delay period to account for the variation.
11. The method of claim 10 , wherein the offline modeling process comprises the controller applying a regression algorithm to simulated host memory buffer access times for different workload classes to estimate a desired host memory buffer access timing.
12. The method of claim 11 , further comprising applying the regression algorithm to received host response latency data.
13. The method of claim 10 , wherein the non-volatile memory system comprises a three-dimensional memory array.
14. A non-volatile memory system comprising:
an interface configured to place the non-volatile memory system in communication with a host comprising a host memory buffer and a command queue:
a non-volatile memory; and means, in communication with the non-volatile memory and the interface, for:
determining, in an offline modeling process, a first delay period and a second delay period
identifying a workload class for a set of commands in the command queue on the host to be executed by a controller of the non-volatile memory system, second delay period;
adding the first delay period to accesses made by the controller to a first memory region of the host memory buffer to execute the set of commands;
adding the second delay period to accesses made by the controller to a second memory region of the host memory buffer to execute the set of commands;
receiving host response latency data from the host in response to a variation between a time spent executing the set of commands and an expected time to execute the set of commands; and
in response to receiving the host response latency data, modifying one or both of the first delay period and second delay period to account for the variation.
15. The non-volatile memory system of claim 14 , wherein the means for determining comprises a controller configured to apply a regression algorithm to simulated host memory buffer access times for different workload classes.
16. The non-volatile memory system of claim 14 , wherein the means for modifying comprises a controller configured to apply a regression algorithm to received host response latency data.
17. The non-volatile memory system of claim 16 , wherein the host memory buffer comprises a portion of a volatile memory on the host under control of the controller of the non -volatile memory system.
18. The non-volatile memory system of claim 14 , wherein the non-volatile memory comprises a three-dimensional memory array.