WO2015196416A1

WO2015196416A1 - Data storage method and device and non-volatile memory

Info

Publication number: WO2015196416A1
Application number: PCT/CN2014/080842
Authority: WO
Inventors: 沙行勉; 石亮; 朱冠宇; 王元钢; 诸葛晴凤
Original assignee: 华为技术有限公司
Priority date: 2014-06-26
Filing date: 2014-06-26
Publication date: 2015-12-30
Also published as: CN105408961B; CN105408961A

Abstract

Embodiments of the present invention provide a data storage method and device and a non-volatile memory (NVM). An NVM controller divides a to-be-written data block and a written-back data block, compares each sub-data block of the to-be-written data block with each sub-data block of the written-back data block to obtain the write power consumption of each sub-data block of the to-be-written data block, determines a write relationship corresponding to the minimum write power consumption of the to-be-written data block according to the write power consumption of each sub-data block of the to-be-written data block, and writes the sub-data blocks of the to-be-written data block into an NVM according to the write relationship corresponding to the minimum write power consumption. By means of the method, the correspondence of the minimum write power consumption is selected by means of comparison, the sub-data blocks of the to-be-written data block are written into the NVM, and thereby the write power consumption of the NVM can be lowered to the lowest level.

Description

Data storage method, device and nonvolatile memory

Technical field

Embodiments of the present invention relate to data communication technologies, and in particular, to a data storage method, apparatus, and nonvolatile memory. Background technique

Non-Volatile Memory (NVM) is characterized by no loss of data stored during a power outage. Common NVMs include Flash Memory, Phase Change Memory (PCM), and Fe Random Access Memory (FRAM). In NVM applications, whether NVM's reset (RESET) operation (ie write 1 operation) or set (SET) operation (ie write 0 operation), NVM's single write energy consumption is higher than read energy consumption 10 -100 times or so, and NVM needs to continuously perform write operations or erase operations during work. Therefore, reducing the number of NVM writes can reduce the energy consumption of the NVM system.

The prior art proposes a Flip-N-Write method for reducing the number of writes to memory. In the method, the memory controller first reads the data to be written from the memory into the register through a read operation, and then performs a write-back data byte of the data to be written and a write-back data byte of the write-back data by bit. XOR operation. Suppose the size of one byte is N bits. If the number of 0s in the XOR result is greater than N/2 (that is, the write back data byte is mostly the same as the data byte to be written), the memory controller updates the data word to be written. Not the same data in the section. If the number of 0s in the XOR result is less than or equal to N/2 (that is, the write back data byte is mostly different from the data byte to be written), the memory controller first flips the data byte to be written, and is about to be written. The data in the data byte is changed from 0 to 1 or from 1 to 0. Then, the memory controller updates the same data in the data byte to be written, because the same data in the data byte to be written is more than different. Data, further reduced the number of updates. In this method, an identifier bit is also added for each byte, and the flag bit is used to indicate whether the data of the corresponding byte is inverted, for example, 0 is used to indicate flipping, and 1 is not inverted, when the data to be written is written into the memory. Write the identifier of the byte to the end of the byte. When the byte is read from the memory, it can be determined according to the identifier of the byte whether the data needs to be flipped for the read byte. Bytes were data when they were stored Flip, then you need to flip the data of the read bytes to restore the data correctly.

The prior art solution can reduce the number of writes of the NVM to a certain extent, thereby reducing the power consumption of the NVM, but each byte in the scheme requires an additional one bit to restore the data, and the additional information accounts for a smaller proportion of storage. Large, when applying the prior art method to external storage, a large amount of data will take up more additional storage space. Summary of the invention

Embodiments of the present invention provide a data storage method, apparatus, and non-volatile memory, which can reduce write energy consumption of an NVM.

A first aspect of the present invention provides a data storage method, the method being applied to a non-volatile memory NVM, the method comprising:

Receiving a write request, where the write request includes a data block to be written and an address;

And respectively, the data block to be written and the write back data block corresponding to the address in the cache are equally divided into N sub-blocks, where N is a positive integer not less than 2;

Comparing each of the sub-blocks of the data block to be written with the sub-block of each of the write-back data blocks to obtain a write energy consumption of each sub-block of the data block to be written; Determining, according to a write energy consumption of each sub-block of the data block to be written, a write relationship corresponding to a minimum write energy consumption of the data block to be written, where the write relationship includes the to-be-written Corresponding relationship between the sub-block of the data block and the sub-block of the write-back block;

The sub-block of the data block to be written is written to the NVM according to the address and the write relationship.

With reference to the first aspect of the present invention, in a first possible implementation manner of the first aspect of the present invention, the sub-data block of each of the data blocks to be written and the sub-data block of each of the data blocks are written The data blocks are compared to obtain the write energy consumption of each of the sub-blocks of the data block to be written, including:

Comparing each of the sub-blocks of the data block to be written with the sub-block of each of the write-back data blocks, respectively, to obtain each sub-block of the data block to be written in each comparison process Writing 0 times and writing 1 times, wherein each of the comparing processes includes a process of comparing one sub-block of the data block to be written with one sub-block of the write-back data block; Write each sub-block of the data block 0 times during each comparison The number, the number of writes, and the preset single write 0 energy consumption and the single write 1 energy consumption calculate the write energy consumption of each sub-block of the data block to be written in each comparison process.

With reference to the first aspect of the present invention, and the first possible implementation manner of the first aspect, in a second possible implementation manner of the first aspect of the present disclosure, The write energy consumption determines a write relationship corresponding to the minimum write energy consumption of the data block to be written, including:

Constructing a bipartite graph with each sub-block of the data block to be written and each sub-block of the write-back data block as a vertex, and each edge weight of the bipartite graph is a corresponding to be written in the side The write power consumption of the sub-block of the data block in a comparison process, wherein the write data energy of the sub-block of the data block to be written in one comparison process is the sub-data block to be written Writing a write energy to a data block of the write back data block;

Determining a best match of the bipartite graph, the best match satisfies the following condition: the sum of weights of each side is the smallest, covering all the vertices of the bipartite graph, and any two sides have no common vertices; The matching is the writing mode corresponding to the minimum write energy consumption.

In conjunction with the first aspect of the present invention and the first and second possible implementations of the first aspect, in a third possible implementation of the first aspect of the present invention, the Writing the sub-block of the data block to be written to the NVM includes: calculating, according to the address and the size of each sub-block of the write-back data block, each sub-data of the write-back data block The address of the block;

The sub-block of the block to be written is written to the NVM according to the write relationship and the address of each sub-block of the write-back data block.

With reference to the first aspect of the present invention, and the first to third possible implementation manners of the first aspect, in a fourth possible implementation manner of the first aspect of the present invention, After the relationship is written to the NVM by the sub-block of the data block to be written, the method further includes:

Writing the write relationship to the NVM to recover the sub-data block to be written from the NVM, and recovering the to-be-written data block according to the write relationship.

A second aspect of the present invention provides a data storage device, including:

a receiving module, configured to receive a write request, where the write request includes a data block to be written and an address; a dividing module, configured to respectively divide the data block to be written and the write back data block corresponding to the address in the cache into N sub-blocks, where N is a positive integer not less than 2; Comparing the sub-blocks of each of the data blocks to be written with the sub-blocks of each of the write-back data blocks to obtain write energy consumption of each sub-block of the data block to be written ;

a determining module, configured to determine, according to a write energy consumption of each sub-block of the data block to be written, a write relationship corresponding to a minimum write energy consumption of the data block to be written, where the write relationship Corresponding relationship between the sub data block of the data block to be written and the sub data block of the write data block;

And a write module, configured to write the sub-block of the data block to be written into the non-volatile memory NVM according to the address and the write relationship.

With reference to the second aspect of the present invention, in a first possible implementation manner of the second aspect of the present disclosure, the acquiring module is specifically configured to:

Comparing each of the sub-blocks of the data block to be written with the sub-block of each of the write-back data blocks, respectively, to obtain each sub-block of the data block to be written in each comparison process Writing 0 times and writing 1 times, wherein each of the comparing processes includes a process of comparing one sub-block of the data block to be written with one sub-block of the write-back data block; Each sub-block of the write data block calculates the number of writes 0, the number of writes 1 and the preset single write 0 power consumption and the single write 1 power consumption in each comparison process to calculate each of the data blocks to be written The write energy consumption of each sub-block in each comparison process.

In conjunction with the second aspect of the present invention and the first possible implementation of the second aspect, in a second possible implementation manner of the second aspect of the present disclosure, the determining module is specifically configured to:

Determining a best match of the bipartite graph, the best match satisfies the following condition: the sum of weights of each side is the smallest, covering all the vertices of the bipartite graph, and any two sides have no common vertices; The matching is the writing mode corresponding to the minimum write energy consumption. With reference to the second aspect of the present invention and the first and second possible implementation manners of the second aspect, in a third possible implementation manner of the second aspect of the present invention, the writing module is specifically configured to:

Calculating an address of each of the sub-data blocks of the write-back data block according to the address and a size of each sub-block of the write-back data block;

With reference to the second aspect of the present invention and the first to third possible implementation manners of the second aspect, in a fourth possible implementation manner of the second aspect of the present invention, the writing module is further configured to: Writing a write relationship to the NVM to recover the sub-data block to be written from the NVM, and recovering the to-be-written data block according to the write relationship.

A third aspect of the present invention provides a nonvolatile memory NVM, including: a storage medium, an NVM controller, and a communication interface;

The communication interface is configured to receive a write request;

The NVM controller is configured to perform the method provided by the first aspect of the present invention and the first to fourth possible implementations of the first aspect;

The storage medium is configured to store data.

In the data storage method, device and non-volatile memory of the embodiment of the present invention, the NVM controller divides the data block to be written and the data block to be written back, by using each sub-block of each data block to be written Writing back the sub-blocks of the data block for comparison, obtaining the write energy consumption of each sub-block of the data block to be written, and determining the write-to-write according to the write energy consumption of each sub-block of the data block to be written The write relationship corresponding to the minimum write power consumption of the data block, and the sub data block to be written to the data block is written to the NVM according to the write relationship corresponding to the minimum write power consumption. The method writes the sub-blocks of the data block to be written into the NVM by comparing the correspondences of the minimum write energy consumption, so that the write power consumption of the NVM can be minimized.

DRAWINGS

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, a brief description of the drawings used in the embodiments or the prior art description will be briefly made below.

1 is a schematic structural diagram of a hardware of a computer according to an embodiment of the present invention;

2 is a schematic structural diagram of hardware of another computer according to an embodiment of the present invention; FIG. 3 is a flowchart of a data storage method according to an embodiment of the present invention; FIG. 4 is a schematic diagram of a write relationship between a data block to be written and a write data block; FIG. a flow chart for determining a write relationship corresponding to a minimum write energy consumption;

FIG. 6 is a schematic structural diagram of a data storage device according to an embodiment of the present disclosure;

FIG. 7 is a schematic structural diagram of an NVM according to an embodiment of the present invention. detailed description

The technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. It is a partial embodiment of the invention, and not all of the embodiments.

The method of the embodiment of the present invention is described by taking a scenario in which the NVM is used as an external memory. FIG. 1 is a schematic diagram of a hardware structure of a computer according to an embodiment of the present invention. As shown in FIG. 1 , the computer includes: a motherboard 11 and a central portion. Central Processing Unit (CPU) 12, Memory 13, NVM14 and external memory 15. The CPU 12, the memory 13 and the NVM 14 are integrated on the main board 11, and the CPU 12 and the memory 13 are connected and communicated via a bus. The commonly used NVMs include PCM, Flash, FRAM, and NAND. The NVM 14 includes: a storage medium, an NVM controller, and a communication interface. The storage medium is used to store data, and the NVM controller can pass a Field-Programmable Gate Array (Field-Programmable Gate Array, Referred to as FPGA) or Application Specific Integrated Circuit (ASIC), the NVM controller has the processing power. The memory 13 is connected to the NVM through an input/output (I/O) interface. The I/O interface can be a Serial ATA interface (SATA) or a Serial SCSI interface (Serial Attached SCSI). SAS), PCI Express (PCI Express). It can be understood that other hardware such as a cache and a south bridge chip can be integrated on the main board 11, which is not enumerated in the embodiment of the present invention, and is not limited thereto.

FIG. 2 is a schematic diagram of a hardware structure of another computer according to an embodiment of the present invention. As shown in FIG. 2, the computer of this embodiment includes: a motherboard 21, a CPU 22, a memory 23, and an NVM 24. The CPU 22 and the memory 23 are integrated in the motherboard. On the 21st, the NVM24 is external to the motherboard 21, the NVM24 is connected to the motherboard through an I/O interface, the NVM 24 includes an NVM controller, and the NVM 24 can be made up. The expansion card is connected to the motherboard 21 through the PCIe interface. When there are multiple NVMs, the expansion card can be connected to the motherboard through multiple expansion cards. Compared with the computer shown in FIG. 1, the NVM in the computer of this embodiment is not integrated on the motherboard but outside the motherboard. In this embodiment, the motherboard 21 can also integrate other hardware such as a cache and a south bridge chip, which are not enumerated here.

FIG. 3 is a flowchart of a data storage method according to an embodiment of the present invention. The method of the embodiment of the present invention may be performed by an NVM controller in the computer shown in FIG. 1 and FIG. As shown in FIG. 3, the method in this embodiment may include the following steps:

Step 101: The NVM controller receives a write request, where the write request includes a data block to be written and an address.

When the NVM controller receives the write request sent by the CPU, it checks whether there is a writeback data block corresponding to the address in the buffer according to the address included in the write request, and if the write data corresponding to the address is in the buffer area, the execution is performed. Step 102: If there is no write-back data block corresponding to the address in the buffer area, the NVM controller reads the write-back data block corresponding to the address from the NVM according to the address. Normally, the NVM controller can read back the data block from the NVM. When the computer has a hardware failure or a write conflict, the NVM controller cannot read back the data block from the NVM. When the NVM controller cannot read back the data block from the NVM, the NVM controller can directly write the data block to be written back to the NVM.

It should be noted that, in the embodiment of the present invention, the data carried in the write request is referred to as a data block to be written, and the data buffered in the buffer area and the same position in the NVM is referred to as a write back data block. Here, both the data block to be written and the data block to be written back refer to the data.

Step 102: The NVM controller divides the data block to be written and the write data block corresponding to the address in the cache into N sub-blocks, where N is a positive integer not less than 2.

The number of blocks N can be adjusted according to actual needs. N is a positive integer not less than 2, and the size of the data block to be written and the data block to be written back are the same. The data block to be written back and the data block to be written are equally divided into N. After each sub-block, the size of each sub-block to be written is the same, the size of the sub-block of each write-back block is the same, and the sub-block of the block to be written and the sub-block of the write-back are written. The size of the sub-blocks is the same. Assuming that the size of the data block to be written is _{block_S1Ze} , the size of the sub-block of the data block to be written is _{block_size} /N, and the size of the sub-block of the data block to be written back is also block_size/N .

Step 103: The NVM controller writes each sub-block of the data block to be written with each The comparison is performed on the sub-blocks of the block to obtain the write energy consumption of each sub-block of the block to be written.

Assuming that each sub-block of data to be written is recorded as i=l~N, the sub-block of each write-back block is marked as ζ,"= ··Ν, when ι=1, NVM control The first sub-block of the data block to be written is compared with the N sub-blocks of the data block to be written back, and there are a total of N comparison processes, and the first sub-data of the block to be written is obtained by N comparison processes. The write energy consumption of the block during N comparisons, therefore, the first sub-block of the data block to be written has a total of N write energy consumption. Then, the N write energy consumptions of the 1+1th to Nth subblocks are respectively calculated according to the above method for calculating the N write energy consumption of the first subblock. Thereby, the write energy consumption of each sub-block of the data block to be written is obtained, and each sub-block of the data block to be written has N write energy consumption.

The NVM controller compares each sub-block of the data block to be written with the sub-block of each write-back data block to obtain the write energy consumption of each sub-block of the data block to be written, specifically: Comparing each sub-block of the data block to be written with the sub-block of each write-back data block, respectively, to obtain the number of writes 0 and write of each sub-block of the data block to be written in each comparison process. 1 times, wherein each of the comparing processes includes a process of comparing one sub-block of the data block to be written with one sub-block of the write-back data block; and then, according to each block to be written Each sub-block of data in each comparison process writes 0 times, writes 1 times, and presets a single write 0 energy consumption and a single write 1 energy consumption to calculate each sub-block of the data block to be written in each comparison process. Write energy consumption in .

For example, when N is 4, the to-be-written data block and the write-back data block are respectively divided into 4 sub-blocks, and the write energy consumption of the 1st sub-block to be written into the data block includes 4 comparison processes. Write power consumption, the NVM controller first determines the first sub-block of the data block to be written and writes the number of writes 0 and writes 1 of the first sub-block of the write-back data block, specifically: NVM controller comparison The first sub-block of the data block to be written and the data of the corresponding bit of the first sub-block of the write-back data block, if the data on the corresponding bit is the same, the data on the bit is not updated, if corresponding If the data on the bit is different, the data on the bit needs to be updated, and it is determined whether the data on the bit needs to be written to 0 or written to 1 to obtain the first block to be written in the first comparison process. Write 0 times and write 1 times of 1 sub-block.

Since the single write 0 energy consumption and the single write 1 energy consumption are different, in this embodiment, the number of write 0 times and the write 1 times need to be counted, and then the total energy consumption of writing 0 and the total energy consumption of writing 1 are calculated. Out After writing the total energy consumption of 0 and writing the total energy consumption of 1, the total energy consumption of writing 0 and the total energy consumption of writing 1 are added together to obtain the writing energy consumption. The single write 0 energy consumption and the single write 0 energy consumption are fixed values, which are set in advance when the data is initialized. After determining the number of writes 0 and the number of writes 1, the number of times written by 0 is multiplied by the number of writes 0 to get written. The total energy consumption of 0 is multiplied by the number of writes per write by 1 to get the total energy consumption of 1 write.

The NVM sequentially calculates the write energy consumption of the first three sub-blocks of the data block to be written and the other three sub-blocks of the write-back data block by the above method, and the write energy of each sub-block of the data block to be written The consumption includes the write energy consumption of the 4 comparison processes.

Step 104: The NVM controller determines, according to a write energy consumption of each sub-block of the data block to be written, a write relationship corresponding to a minimum write energy consumption of the data block to be written, where the write relationship includes a write relationship to be written. The correspondence between the sub-block of the data block and the sub-block of the write back data block.

4 is used as an example for illustration. FIG. 4 is a schematic diagram of a write relationship between a data block to be written and a write data block. As shown in FIG. 4, the data block to be written is divided into four sub-blocks: d ₂ , d ₃ , d ₄ , the addresses of the 4 sub-blocks to be written to the data block are consecutive, and the write-back data block is also divided into 4 sub-blocks: fl, f2, f3, f4, and 4 of the data block are written back. The address of each sub-block is continuous, and the address here continuously means that the physical address is continuous.

In the prior art, the write relationship between the data block to be written and the write back data block is fixed, that is, dl is written to fl, d2 is written to f2, d3 is written to f3, and d4 is written to f4. In this embodiment, there is no fixed write relationship between the sub-block of the data block to be written and the sub-block of the data block to be written back. When the number of blocks is N, there are a total of N! In the example shown in Figure 4, the value of N is 4. Therefore, there are 24 write relationships. When the data blocks to be written have different write relationships, the write energy consumption is different. The purpose of this step is to write from 24 types. In the relationship, select a write relationship with the lowest write energy consumption.

In an implementation manner, the NVM controller can calculate the write energy consumption corresponding to each write relationship, and compare the write power consumption corresponding to the various write relationships, and select the write with the lowest power consumption. relationship. In another implementation, the NVM controller can determine the write relationship corresponding to the minimum write energy consumption through the bipartite graph. For example, the write relationship corresponding to the minimum write power determined by the NVM controller is: dl→f3, d2→f2, d3→f4, d4→fl, that is, the sub-block dl of the data block to be written is written back to the data block. The sub-data block f3, the sub-data block d2 of the data block to be written is written back to the sub-data block Ω of the data block, and the sub-data block d3 of the data block to be written is written into the sub-data block f4 of the write-back data block, to be written The sub-data block d4 of the incoming data block is written to the sub-data block fl of the write-back data block.

Step 105: The NVM controller records the number of sub-blocks to be written according to the address and the write relationship. The block is written to the NVM.

Specifically, the NVM controller calculates an address of each sub-block of the write-back data block according to the address and the size of each sub-block of the write-back data block; and then, according to the write relationship and the write-back of each sub-block of the data block The address writes the sub-block of the data block to be written to the NVM. When the NVM controller writes the sub-blocks of the data block to be written to the NVM, there are two ways: In the first way, the NVM controller writes back corresponding data according to the sub-blocks of the data block to be written and the write relationship. The sub-block of the data block is overwritten; then, the sub-block of the rewritten write back data block is written into the NVM according to the address of each sub-block of the write-back data block. In the second mode, the NVM overwrites the data in the NVM according to the address and write relationship of each sub-block of the write-back data block.

The NVM controller can calculate the address of each sub-block of the write-back data block according to the following formula: A _r a+j xM, where Α is the address of the sub-block of the first write-back data block, j=l, ..., N, a is the starting address of the address carried in the write request, and M is the size of each sub-block of the write-back data block. After determining the address of each sub-block of the write data block, the NVM controller writes the sub-block of the block to be written to the NVM according to the write relationship.

In this embodiment, the NVM controller divides the to-be-written data block and the write-back data block into blocks, by comparing each sub-block of the data block to be written with each sub-block of the data block that is written back, Obtaining the write energy consumption of each sub-block of the data block to be written, and determining the write corresponding to the minimum write energy of the data block to be written according to the write energy consumption of each sub-block of the data block to be written Relationship, the sub-data block to be written to the data block is written to NVMo according to the write relationship corresponding to the minimum write power consumption. The method compares the correspondence between the selected write energy consumption and the sub-data to be written into the data block. The block is written to the NVM to minimize the write power consumption of the NVM.

Generally, there is a great similarity between the data block to be written and the data block to be written back. When a new data block is inserted or the data block is shifted, causing the address of the data block to be changed back, if the prior art solution is adopted, When the data to be written and the data at the corresponding position of the write data block are compared, most of the data is different. In this case, the write power consumption of the NVM is large. With the method of the embodiment, this type of problem can be effectively solved. Still taking FIG. 4 as an example, when a new data block is inserted between fl and Ω, the addresses of the original f2, f3, and f4 all move backward by the size of one data block. If the prior art scheme is used, dl is written to the existing fl, d2 is written to the existing f2, d3 is written to the existing f3, and d4 is written to the existing f4, due to the original Ω data. Moved backwards, guide The data of d2 and the existing Ω are very different. Correspondingly, the data of d3 is very different from the data of the existing f3. The data of d4 and the existing f4 are also very different. Most of the data needs to be rewritten, NVM. Write energy is very high. In the solution of this embodiment, by comparing the write energy consumption corresponding to various write relationships, dl can be written into the existing fl, d2 is written into the existing f3, and d3 is written into the existing f4, D4 writes the existing f2, since the data of dl and the existing fl are mostly the same, the data of d2 and the existing f3 are mostly the same, and the data of d3 is mostly the same as the data of the existing f4, only a small part of the data is required. Write, therefore, the method of the present embodiment can reduce the write power consumption of the NVM.

After the NVM controller writes the sub-block of the data block to be written to the NVM, the write relationship is also written to the NVM, so as to read the sub-block of the data block to be written from the NVM, and resume the recovery according to the write relationship. Write a block of data. This is because the order of the contents of the data block to be written is disturbed after the data block to be written is written to the NVM according to the method of the embodiment, and therefore, when the data to be written is read from the NVM, Restore the order in which the data blocks are to be written. The write relationship can be stored in a dedicated area in the NVM. When the NVM controller reads data from the NVM, it also needs to read the write relationship corresponding to the data from the dedicated area. The write relationship stored in the NVM may specifically be an identifier of a sub-block of the data block to be written. When N is 2 ^k , the additional overhead required for each sub-block of the data block to be written is log(N) Bit, the overhead refers to the size of the storage area required to store the sub-block of the data block to be written, then the overhead of the data block to be written is N*lo _g (N) bits, when N is not 2 ^K , The additional overhead required for each sub-block of data block to be written is "i ₀ g(N;n, ", indicating an up-rounding operation, and the overhead required to write the data block is N"log(Nfl. For example N=4, 00, 01, 10, 1 1 can be used to represent the four sub-blocks of the data block to be written respectively. The method of this embodiment requires an additional N*l _0g (N;) to be written to the data block. The bit stores the write relationship, whereas in the prior art, each byte requires an extra bit to store the flip information, and therefore, the method of the embodiment can also reduce the overhead.

FIG. 5 is a flowchart of determining a write relationship corresponding to a minimum write power consumption according to an embodiment of the present invention. This embodiment is a detailed description of step 104 in the embodiment shown in FIG. 3, which is adopted in this embodiment. The bipartite graph is constructed to determine the write relationship corresponding to the minimum write energy consumption. The bipartite graph is only a feasible method for solving the write relationship corresponding to the minimum write energy consumption, and can of course be solved by other methods. As shown in FIG. 5, the method in this embodiment may include the following steps:

Step 201: The NVM controller constructs a bipartite graph with each sub-block of the data block to be written and each sub-block of the write data block as a vertex, and the weights of the sides of the bipartite graph are corresponding to each side. Write energy consumption of a sub-block of a data block written in a comparison process, wherein a sub-data block of the data block to be written has a write energy consumption in a comparison process as a sub-data of the data block to be written The block writes the write energy of a sub-block of the write back data block.

Suppose the bipartite graph is represented by G(D, F), D represents the data block to be written, F represents the write back data block, and the vertex of the bipartite graph is the sub data block of the data block to be written or the sub data block of the write data block. Assuming that the write energy consumption of the first sub-block 4 of the data block to be written to the first sub-block of the write-back data block is (1^), the weight of each side of the bipartite graph G is eCd. ,, fj), the bipartite graph can be represented by a matrix of NX N: E = [E ₁ , _j ] _NXN , i = 2..., N}, j = 2..., N}.

Step 202: The NVM controller determines the best match of the bipartite graph. The best match satisfies the following conditions: The sum of the weights of each side is the smallest, covering all the vertices of the bipartite graph, and any two sides have no common vertices.

The existing matching relationship can be used to solve the best matching of the bipartite graph. The explanation is not explained too much. In this embodiment, the best matching needs to satisfy the following three conditions simultaneously: (1) The sum of the weights of each side is the smallest, The weight of each edge is a write energy consumption, and the minimum sum of the weights on each side means that the write energy consumption of each side is the smallest. (2) Covering all vertices of the bipartite graph, since the vertices of the bipartite graph represent sub-blocks of the data block to be written or sub-blocks of the data block to be written back, all vertices covering the bipartite graph are used to ensure that the data block to be written is to be written All of the sub-blocks are written back to the corresponding sub-block of the write-back block. (3) There is no common vertex on any two sides. This is because when a sub-block of the data block to be written is written back to NVM, any sub-block of the block to be written can only be written to the write-back block. If one of the two sub-blocks has a common point, then the sub-block of the data block to be written may correspond to the sub-block of the two data blocks.

Step 203: The NVM controller uses the best match as the write mode corresponding to the minimum write power consumption. FIG. 6 is a schematic structural diagram of a data storage device according to an embodiment of the present invention. As shown in FIG. 6, the device in this embodiment may include: a receiving module 31, a dividing module 32, an obtaining module 33, a determining module 34, and a writing module 35. .

The receiving module 31 is configured to receive a write request, where the write request includes a data block to be written and an address;

The dividing module 32 is configured to divide the data block to be written and the write back data block corresponding to the address into N sub-blocks, wherein N is a positive integer not less than 2; An obtaining module 33, configured to compare the sub-block of each of the data blocks to be written with the sub-block of each of the write-back data blocks to obtain each sub-block of the data block to be written Write energy consumption;

a determining module 34, configured to determine, according to a write energy consumption of each sub-block of the data block to be written, a write relationship corresponding to a minimum write energy consumption of the data block to be written, where the writing The relationship includes a correspondence between the sub-block of the data block to be written and the sub-block of the write-back data block;

The writing module 35 is configured to write the sub-block of the data block to be written into the non-volatile memory NVM according to the address and the write relationship.

In this embodiment, the obtaining module 33 is specifically configured to: compare each of the sub-blocks of the data block to be written with the sub-block of each of the data blocks, to obtain each comparison process. Writing 0 times and writing 1 times of each sub-block of the data block to be written, wherein each of the comparing processes includes one sub-block of the to-be-written data block and the write-back data block a process of comparing sub-blocks; then, according to each sub-block of the data block to be written, the number of writes 0, the number of writes 1 and the preset single write 0 energy consumption and single time in each comparison process Write 1 energy consumption calculates the write energy consumption of each sub-block of the data block to be written in each comparison process.

Optionally, the determining module 34 is specifically configured to: first, construct a bipartite graph with each sub-block of the to-be-written data block and each sub-block of the write-back data block as a vertex, and the bipartite graph The weight of each side of the edge is the write energy consumption of the sub-block of the data block to be written corresponding to each side in a comparison process, wherein the sub-block of the data block to be written is in a comparison process The write power consumption writes the write energy consumption of one sub-block of the write-back data block to the sub-block of the data block to be written. Then, the best match of the bipartite graph is determined, and the optimal match satisfies the condition that the sum of the weights of the sides is the smallest, covering all the vertices of the bipartite graph, and any two sides have no common vertices. Finally, the best match is used as the write mode corresponding to the minimum write energy consumption.

In this embodiment, the writing module 35 is specifically configured to: calculate an address of each sub-block of the write-back data block according to the address and a size of each sub-block of the write-back data block; The write relationship and the address of each sub-block of the write-back data block write the sub-block of the block to be written to the NVM. The writing module 35 is further configured to: Writing a write relationship to the NVM to recover the sub-data block to be written from the NVM, and recovering the to-be-written data block according to the write relationship.

The data storage device of this embodiment may be used to implement the technical solution of the method embodiment shown in FIG. 3, and the implementation principle and technical effects are similar, and details are not described herein again.

FIG. 7 is a schematic structural diagram of an NVM according to an embodiment of the present invention. As shown in FIG. 7, the NVM 400 of the embodiment includes: a storage medium 41, an NVM controller 42 and a communication interface 43. The NVM communicates with other hardware in the computer, and is specifically configured to receive a write request. The NVM controller 42 is configured to perform the data storage method provided by the foregoing embodiment, and the implementation principle and technical effects are similar, and are not described herein again; 42, used to store data. The NVM controller may specifically be a chip having a processing function, such as a CPU, an FPGA, an ASIC, or the like.

A person skilled in the art can understand that all or part of the steps of implementing the above method embodiments may be implemented by hardware related to program instructions, and the foregoing program may be stored in a computer readable storage medium, when executed, The foregoing steps include the steps of the foregoing method embodiments; and the foregoing storage medium includes: a medium that can store program codes, such as a ROM, a RAM, a magnetic disk, or an optical disk.

Finally, it should be noted that the above embodiments are only for explaining the technical solutions of the present invention, and are not intended to be limiting thereof; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art will understand that The technical solutions described in the foregoing embodiments may be modified, or some or all of the technical features may be equivalently replaced; and the modifications or substitutions do not deviate from the technical solutions of the embodiments of the present invention. range.

Claims

Claim

A data storage method, the method being applied to a non-volatile memory NVM, wherein the method comprises:

2. The method according to claim 1, wherein the comparing the sub-blocks of each of the data blocks to be written with the sub-blocks of each of the write-back data blocks to obtain a The write energy consumption of each sub-block of the write data block, including:

The method according to claim 1 or 2, wherein the determining the minimum write energy of the data block to be written according to the write energy consumption of each sub-block of the data block to be written The corresponding write relationship, including:

The method according to any one of claims 1 to 3, wherein the sub-block of the data block to be written is written into the NVM according to the address and the write relationship. , including:

The method according to any one of claims 1 to 4, wherein the sub-block of the data block to be written is written into the NVM according to the address and the write relationship. After that, it also includes:

6. A data storage device, comprising:

a receiving module, configured to receive a write request, where the write request includes a data block to be written and an address;

a dividing module, configured to respectively divide the data block to be written and the write back data block corresponding to the address in the cache into N sub-blocks, where N is a positive integer not less than 2; Comparing the sub-blocks of each of the data blocks to be written with the sub-blocks of each of the write-back data blocks to obtain write energy consumption of each sub-block of the data block to be written ;

The device according to claim 6, wherein the obtaining module is specifically used for:

The device according to claim 6 or 7, wherein the determining module is specifically configured to:

The apparatus according to any one of claims 6-8, wherein the write module is specifically configured to:

The device according to any one of claims 6 to 9, wherein the writing module is further configured to: write the write relationship to the NVM, to read from the NVM When the sub-block of the data block is written, the block to be written is restored according to the write relationship.

11. A non-volatile memory NVM, comprising: a storage medium, an NVM controller, and a communication interface; The communication interface is configured to receive a write request;

The NVM controller is configured to perform the method of any one of claims 1-5; the storage medium is configured to store data.