Storage system and method for prediction-based pre-erase of blocks to improve sequential performance
A storage system erases blocks of memory prior to writing data to the blocks. Instead of erasing the blocks at the time the write operations are executed, the storage system pre-erases the blocks, which can improve performance. However, because program failure errors can occur if the blocks sit empty for a relatively-long period of time prior to programming, the storage system pre-erases the blocks upon a prediction that a host will send sequential write commands to the storage system that will use the blocks. Additionally or alternatively, the storage system can pre-erase a block upon determining that the number of write commands in a command queue in the storage system is above a threshold that represents a number of write commands needed to fill the block with data.
1. A storage system comprising:
a memory; and
a controller configured to:
perform garbage collection on a plurality of blocks in the memory;
predict that a host will send sequential write commands to the storage system;
prior to receiving the sequential write commands from the host, pre-erase the plurality of blocks in the memory, wherein pre-erasing the plurality of blocks comprises performing erase operations on the plurality of blocks that would otherwise be performed as part of write operations of the plurality of blocks;
receive the sequential write commands from the host; and
use the pre-erased plurality of blocks to store data written by the sequential write commands.
2. The storage system of claim 1 , wherein the controller is further configured to store the data written by the sequential write commands as a foreground operation, and wherein because the plurality of blocks are pre-erased, the controller is further configured to avoid performing an erase operation on the plurality of blocks during the foreground operation.
3. The storage system of claim 1 , wherein the controller is further configured to use a machine-learning model to predict that the host will send the sequential write commands to the storage system.
4. The storage system of claim 3 , wherein inputs to the machine-learning model comprise at least one of the following items: a logical block address, a length of logical block addresses, a timestamp, a frequency of host writes, and a frequency of host reads.
5. The storage system of claim 1 , wherein the controller is further configured to pre-erase the plurality of blocks during idle time.
6. The storage system of claim 1 , wherein the controller is further configured to store addresses of the pre-erased plurality of blocks in a first-in-first-out (FIFO) buffer.
7. The storage system of claim 6 , wherein the controller is further configured to remove an address of a block from the FIFO buffer in response to determining that the block has been in the buffer more than a threshold amount of time.
8. The storage system of claim 1 , wherein the controller is further configured to:
determine that a number of write commands in a command queue in the storage system is above a threshold; and
in response to determining that the number of write commands in the command queue is above the threshold, pre-erase another block in the memory.
9. The storage system of claim 8 , further comprising:
a front-end module comprising the command queue; and
a back-end module configured to execute the write commands.
10. The storage system of claim 8 , wherein the controller is further configured to pre-erase the another block using a segmented erase process.
11. The storage system of claim 1 , wherein the memory comprises a three-dimensional memory.
12. In a storage system comprising a memory, a method comprising:
performing garbage collection on a block in the memory;
determining that a number of write commands in a command queue in the storage system is above a threshold;
in response to determining that the number of write commands in the command queue is above the threshold, pre-erasing the block in the memory, wherein pre-erasing the block comprises performing an erase operation on the block that would otherwise be performed as part of a write operation of the block; and
executing the write commands, wherein the pre-erased block is used to store data written by the write commands.
13. The method of claim 12 , wherein the storage system comprises a front-end module comprising the command queue and a back-end module configured to execute the write commands.
14. The method of claim 13 , wherein the block in the memory is pre-erased in phases when the front-end module is busy.
15. The method of claim 12 , wherein the threshold represents a number of write commands needed to fill the block with data.
16. The method of claim 12 , further comprising:
predicting that a host will send sequential write commands to the storage system; and
prior to receiving the sequential write commands from the host, pre-erasing a plurality of blocks in the memory that will be used to store data written by the sequential write commands.
17. The method of claim 16 , further comprising using a machine-learning model to predict that the host will send the sequential write commands to the storage system.
18. The method of claim 16 , further comprising pre-erasing the plurality of blocks during idle time.
19. The method of claim 16 , further comprising removing one of the plurality of pre-erased blocks from a pool of available blocks in response to determining that the one of the plurality of pre-erased blocks was pre-erased more than a threshold amount of time ago.
20. A storage system comprising:
a memory;
means for performing garbage collection on first and second blocks in the memory; and
one or both of the following:
means for passive pre-erasing of the first block in the memory based on a machine-learning prediction, wherein passive pre-erasing of the first block comprises performing an erase operation on the first block that would otherwise be performed as part of a write operation of the first block; and
means for active pre-erasing of the second block in the memory based on a command-queue-based prediction, wherein active pre-erasing of the second block comprises performing an erase operation on the second block that would otherwise be performed as part of a write operation of the second block.