JP6034467B2 - system - Google Patents

Description

  Embodiments described herein relate generally to a memory system having a key-value store scheme that is accessed by a host system.

  As a storage device included in a general host system, for example, a computer system, there are a magnetic hard disk drive (HDD), a solid-state drive (SSD) equipped with a nonvolatile semiconductor memory, and the like. Although SSD is classified as storage, it can be said that it is a memory system whose scale and function are expanded.

  The memory system includes, for example, an interface, a first memory block, a second memory block, and a controller. The first memory block stores a file as data, and the second memory block is a buffer memory at the time of data writing / reading. The first memory block is non-volatile and has a larger capacity than the second memory block, but the access speed is slower. The second memory block is used to compensate for the speed difference between the communication speed of the interface and the writing / reading speed of the first memory block. For example, the first memory block is a non-volatile flash memory, and the second memory block is a volatile DRAM or SRAM. Such a conventional storage-type memory system has a configuration for realizing a data write / read function by address designation.

  On the other hand, data such as another text associated with a certain text stored in a memory system, a specific bit pattern in a binary file, a specific pattern in a video file, and a characteristic audio pattern in an audio file can be efficiently used. In order to retrieve the data, it is desirable to have a data read function by data designation. For this reason, not only normal data storage but also a method in which metadata associated with data is attached and stored, and the metadata is referred to in order to obtain desired data.

  Metadata management methods can be broadly divided into two types: a table-type database type and a key-value store (KVS) method in which data correspond one-to-one. In the KVS system, when a key is given as a search request, the value associated with it is output.

  In order to realize the KVS system in the conventional system, after managing the data stored in the memory system and a plurality of metadata in the main memory (DRAM) of the host system, the central processing unit (CPU) ) Is used to repeatedly perform data input / output processing in which data is read from the storage and verified again.

  The key-value store (KVS) method in the conventional system and its problems will be described.

  When the KVS method is realized by the conventional SSD, the data is stored as a file, and the metadata of the key-value type data (or key-value pair) attached to the data is also stored as a file. In other words, the key-value store method is realized by a host system higher than the file system, and is a file system or application implemented in the OS.

  In this case, although there is an advantage that the key-value store method can be realized with a general-purpose hardware configuration, since handling of metadata is the same as normal data, read / write or search operations of metadata are performed by the host. The system reads the metadata file into the main memory (for example, DRAM) and handles it. This causes, for example, at least three problems as follows.

  First, file access performance is degraded. In general, since the size of the main memory is smaller than that of the SSD, a metadata file larger than the main memory size cannot be handled at a time. For this reason, the metadata file is divided for each key, for example, and a metadata file having an easy-to-handle size is read out to the main memory as needed and used. This process is repeated until the required key-value is obtained, and file access corresponding to the number of metadata files occurs for the SSD, so the SSD file access speed is slower than the metadata read request speed. In this case, the entire host system and local system (memory system) are limited.

  Second, the load on the CPU increases. Since creation, management, and collation (search) of metadata are all processed by the CPU, a load is imposed on the CPU during the metadata processing. In particular, since the metadata is created corresponding to the data, when the data is updated, it is necessary to find and update the corresponding metadata from the metadata file. In addition, since a mechanism for searching for metadata must be performed by using a software algorithm in the CPU, the CPU eventually creates a new load for metadata management.

  Third, the load on the bus or interface increases. As a result of the first and second problems, metadata information frequently goes back and forth between the host system and the local system (memory system), increasing the traffic on the bus and the interface unit.

JP 2005-209214 A Japanese Patent No. 4167359

  Provided is a memory system that can efficiently and quickly process an operation request for metadata by a key-value store method.

A system according to an embodiment includes a metadata table including key-value type data that is a set of a key and a value corresponding to the key in a system that receives a command from a host and transfers data to the host via an interface. And a non-volatile memory for storing the data area, and a controller for accessing the non-volatile memory, and the key-value type data is stored in a storage location specified by the first address of the metadata table. The value stored in the storage location specified by the first address stores the second address indicating the storage location in the data area, and the value is stored in the storage location specified by the second address. A third address indicating the storage location and the key are stored, When a command for registering a value associated with the key is input from the host, the controller searches whether the key exists in the metadata table, and the key exists in the metadata table. If not, it is determined whether or not the value associated with the key is stored in the data area. If the value is not stored in the data area, the value is designated by the fourth address of the data area. When the value is stored in the data area, the key and the fourth address storing the value are stored in the storage location specified by the fifth address of the metadata table. Register .

It is a figure which shows typically the relationship between the real data area in a 1st memory block, and a metadata table, and the mechanism of a key-value store system. It is a figure which shows the relationship between the element in FIG. 1, and a set. It is a flowchart which shows the specific process sequence in a key-value store system. It is a flowchart which shows the specific process sequence in a key-value store system. It is a flowchart which shows the specific process sequence in a key-value store system. It is a flowchart which shows the specific process sequence in a key-value store system. It is a block diagram which shows the hardware constitutions of the memory system of 1st Embodiment. It is a block diagram which shows the hardware constitutions of the memory system of 2nd Embodiment. It is a block diagram which shows the hardware constitutions of the memory system of 3rd Embodiment. It is a block diagram which shows the hardware constitutions of the memory system of 4th Embodiment. It is a block diagram which shows the hardware constitutions of the memory system of 5th Embodiment. It is a block diagram which shows the hardware constitutions of the memory system of 6th Embodiment. It is a block diagram which shows the hardware constitutions of the memory system of 7th Embodiment. It is a block diagram which shows the hardware constitutions of the memory system of 8th Embodiment. It is a figure which shows typically the real data area and metadata table of the modification 1. FIG. It is a figure which shows typically the real data area and metadata table of the modification 2. It is a figure which shows typically the real data area and metadata table of the modification 3. It is a figure which shows typically the other realization method of the key-value store system of the modification 3. It is a figure which shows typically the real data area and metadata table of the modification 4.

  In the following description, components having substantially the same function and configuration are denoted by the same reference numerals, and redundant description will be given only when necessary.

  Metadata stored in the memory system is stored using the key-value store method. An embodiment for providing an efficient storage method and configuration of the key-value type data will be described with reference to the drawings.

<Metadata table and key-value store method>
First, a metadata table and a key-value store system, which are basic principles of the present embodiment, will be described.

  FIG. 1 schematically shows the relationship between the actual data area in the first memory block and the metadata table, and the mechanism of the key-value store method. Numerical values and symbols are used for explanation and do not necessarily correspond to the embodiment.

  As shown in FIG. 1, an address-designated real data area (real address space) 161 and a metadata table 162 exist in the physical address space accessible by the memory. The actual data area 161 directly corresponds to the logical address space in the conventional concept. Since the metadata table 162 is a data area that is appropriately used by the memory system side, the user or client does not explicitly access it during normal use. However, explicit access to the system is also permitted as a test command. The user basically sends a request to manipulate the actual data or metadata to the memory system using interface commands, and the memory system returns a signal or data as a response to the user (host system) as a result of internal processing. .

  The memory system distinguishes between storage areas of the actual data area 161 and the metadata table 162 in a logical address-physical address conversion table for converting a logical address into a physical address. Since the metadata address in the metadata table 162 is created as necessary, it may not exist in the first memory block unless there is a request for creating key-value type data.

  As described above, the storage capacity of the metadata table 162 in the memory system according to the present embodiment is not fixed and is arbitrarily set (expanded or reduced) according to a request based on the key-value store method. Can exist. For this reason, the user can use the accessible physical memory space with maximum efficiency while arbitrarily handling the metadata. In extreme terms, it is possible to handle no metadata at all. In this case, the physical memory space can be utilized to the maximum extent.

  Conversely, if the metadata is fully utilized, the metadata table 162 may be enlarged to the same size or more as the actual data area 161. Even in this case, since the management of the metadata is a work on the memory system (local system) side, the host system side is released from the management of the metadata. Therefore, the user (host system side) does not need to be aware of metadata management in normal use.

  Although the key-value pair is stored in the metadata table 162, the actual data exists in the actual data area (real address space) 161.

  The relationship between the metadata table 162 and the actual data in the actual data area 161 in this embodiment will be described with reference to a specific example of data extraction by the key-value store method.

  The key-value store (KVS) system means a database management system in which a set of keys (keys) and values (values) is written, and values can be read by designating the keys. The KVS method is often used over a network, but there is no doubt that the data storage destination is some local memory or storage system. Even metadata is stored in the actual data address space.

  Data is normally read by designating the top address of the stored memory. Here, the data may be rephrased as a file. Depending on the file system, the actual data address space is managed in units of 512-byte sectors, for example. Alternatively, if it is not necessary to limit the file system, for example, it may be managed in units of 4 KB or 8 KB which are the page size of reading / writing of the NAND flash memory.

  The actual data area 161 shown in FIG. 1 conceptually shows how a file (data) corresponding to an actual address is stored. For example, it is assumed that% a-file.txt, which is the file ID of the file a-file.txt, is stored at the actual data address & 001, and the file content “This is a book” is stored at & 002. Actually, since the actual data address is generally managed in units of bytes, it is a special example that it is continuous as & 001 and & 002.

  The metadata table 162 shown in FIG. 1 shows stored key-value pairs. The key is extracted from the data file and is a search target element or set. In this example, the key is a set and includes elements & 001 and & 011. Value is a value returned when a key is found. In this example, a set of file IDs of data files including the word “book” is stored as values in the form of addresses in the actual data address space.

According to the procedure in FIG.
(I) Enter key as entry and search for value from metadata table.
(Ii) Since the value corresponding to the found key is an actual data address that stores the set to which the key belongs, the actual data address is referred to.
(Iii) The data written in the referenced actual data address is output.
It becomes.

  The relationship between the actual data address and the metadata table, and the relationship between the key-value pair is an element-set relationship as shown in FIGS. 2 (a) and 2 (b).

  That is, in a normal file, as shown in FIG. 2 (b), for example, files with a file name “a-file.txt” are a set, and each word of text data “This is a book” is an element. It is. The file ID is also an element.

  On the other hand, when placed in the metadata table (metadata address space), as shown in FIG. 2A, the relationship between the set and the element can be reversed and rearranged. That is, it is converted into a “transposition” relationship. In the set “book”, file names “a-file.txt” and “b-file.txt” are stored as elements. In the key-value type data, the rearranged set name (“book”) is searched and the elements (“a-file.txt”, “b-file.txt”) are obtained. . In general, this is the creation of a transposed file executed in full-text search and the search procedure itself, which can be said to be a practical example of key-value type data.

  An inverted file is an index file for search used in the inverted index method, which is one of the methods for realizing the full-text search function. In the inverted index method, an index data file called an inverted file that stores a list of files containing the content for each content is created in advance, and the contents of the inverted file are updated each time a file is added / deleted or periodically. To do. In response to a content search request, the contents of the transposed file corresponding to the search target content may be output as a search result. Therefore, it is not necessary to check the contents of all files every time a full text search is performed. Therefore, the search can be speeded up. It should be noted that the transposed file is one example of key-value type data, and the present embodiment is not limited to this.

<Key-value store command>
The host system gives the next command for a key-value store request to the host interface of the memory system.

  A command for a request based on the key-value store method includes a command (PUT) for registering a new set (value) associated with the key, and a new element (value) for the set (value) associated with a certain key. A command (WRITE) for adding a command, a command (GET) for storing an element of a set (value) associated with the key in the buffer and returning its size, a command (READ) for reading the element (value) stored in the buffer, including.

  The command name may be changed as appropriate, and a new command may be added for a request based on the key-value store method. For example, a command for organizing the elements (values) belonging to the set (key), a command for rearranging the sets (key) in the metadata table, and comparing the elements (value) with each other.

In the present embodiment, the metadata table and the actual data area are linked in the memory system according to the command request. Specific procedures such as addition and retrieval of key-value type data when the above command is used are shown in the flowcharts of FIGS. 3A to 3D.
(1) When registering new key-value type data (PUT), as shown in FIG. 3A, first, it is searched whether the key already exists in the metadata table (steps S1 and S2). If the key is found, an error is returned to the output, that is, it returns that the key already exists and ends (step S3).

  On the other hand, if the key is not found, the process proceeds to value processing. It is searched whether the value is stored in the actual data area (actual data address) (step S4). If the value is not stored in the actual data area at the time of registration of the key-value type data, the value is added to the actual data area (step S5). If the value is stored, the key is stored in the metadata table as it is, and the actual data address of the value is associated with the key and registered (step S6).

When the logical address-physical address conversion table is managed in the memory system, the management circuit is notified of the update of the metadata table (step S7). Finally, the actual data size of value is output and the process ends (step S8).
(2) When adding a new value to an existing key (WRITE), as shown in FIG. 3B, first, it is searched whether the key exists in the metadata table (steps S11 and S12). If no key is found, for example, size = 0 is returned to notify the output that there is no key (step S13).

  On the other hand, if the key is found, the storage location specified by the actual data address stored in the value corresponding to the key is referred to (step S14), and a new value is added to the storage location. First, it is checked whether or not there is a space in the value storage location (step S15). If there is no space in the storage location of the value, a pointer for jumping to the actual data address of another value is stored (step S16). Subsequently, a new value is added to the storage location designated by the address (step S17).

If there is a vacancy in the storage location of the value, a new value is added to the vacant storage location of the actual data address of the value (step S17). Finally, the actual data size of value is output and the process ends (step S18).
(3) When obtaining a set (value) associated with the key (GET), as shown in FIG. 3C, first, it is searched whether the key exists in the metadata table (steps S21 and S22). If the key is not found, for example, size = 0 is returned to notify the output that the key does not exist (step S23).

On the other hand, if the key is found, the storage location designated by the actual data address stored in the value corresponding to the key is referred to (step S24). Then, the value stored in the storage location of the actual data address is read and stored in the buffer memory or the register memory (step S25). Finally, the actual data size of value is output and the process ends (step S26).
(4) When outputting the elements of a set (value) stored in the buffer memory (or register memory) (READ), as shown in FIG. 3D, in the buffer memory storing the elements of the set (value) Is referred to (step S31). Then, it is checked whether or not there is an element of the set (value) (step S32). If no element of the set (value) is found, for example, S = NULL is returned in the output to notify that the element of the set (value) does not exist (step S33).

  On the other hand, if an element of the set (value) is found, the element of the set (value) corresponding to the designated size is read (step S34). Subsequently, the elements of the read set (value) are output and the process ends (step S35). Here, an example in which an element is read out by size designation has been shown, but actually, the location of the buffer memory may be specified and read out.

  In the procedure (3), there may be a case where the head address of the actual data address of value is returned to the host system. This is because the following procedure (4) is usually performed after this procedure, which is convenient for reading the actual data of value. How to define depends on the definition of the command set. In the present application, a specific command set is used to explain the procedure of the key-value store method, and the present invention is not limited to this. Similarly, other procedures are not limited to the above description.

  When the memory system uses Hash-CAM, there is a possibility that the key and the value do not match, that is, the key and the value are not associated with each other. The Hash-CAM will be described later in detail.

  For this reason, in the case of Hash-CAM, in the procedure for searching for a key in the metadata table, a procedure for determining whether the actual data and the key match with reference to the value associated with the key is added. If they do not match, the search is performed again by designating the metadata address in another key range according to the address management rule of the key-value store method in the Hash-CAM (for example, checking the adjacent address).

  Actual procedures and commands are not limited to this. For example, when a plurality of keys are found in actual operation, it is possible to give variations to the method such as setting a flag once and reading the values together.

  Further, as described above, for example, a command for organizing elements (values) belonging to a set (key), a rearrangement of sets (keys) in the metadata table, and a comparison between elements (values) are given. There can be a procedure for the command to do.

  It should be noted that the key and value can be in a set and the other is an element, or vice versa. Alternatively, since they correspond one-to-one, both may be a set or an element.

  In the present embodiment, a host interface or a local controller or a memory controller via the host interface or the like can receive the search command and execute a series of processes of the key-value store method. In addition, the local controller in the memory system or the memory controller can be provided with a direct memory access controller (DMAC). In this case, the memory system mainly operates in a key-value store system. Can be controlled. In some cases, it is possible to independently access another memory outside the memory system (for example, the main memory of the host system).

  Hereinafter, a specific hardware configuration of this embodiment will be described with reference to the drawings as appropriate.

[First Embodiment]
A hardware configuration of the memory system according to the first embodiment will be described.

  FIG. 4 is a block diagram illustrating a hardware configuration of the memory system according to the first embodiment.

  As shown, the memory system (or local system) 10 includes a host interface 11, a local controller 12, a memory controller (or chip controller) 13, a fixed length data generator 14, a register memory (or cache memory, page register, R / W register, also called page cache) 15, and a first memory block 16.

  For example, a host system is connected to the host interface 11 of the memory system 10 via a bus such as AMBA, SATA, PCIe, or USB. The host system has a CPU 101 and a main memory 102.

  The first memory block 16 stores an actual data area 161 and a metadata table 162 extracted from the actual data area 161. The metadata table 162 has key-value type data.

  The key-value type data in the metadata table 162 stores a key as metadata associated with the data and a head address (value) of the actual data address of the associated data as a list. Using this key-value type data, for example, the above-described transposed file can be configured.

  For the first memory block 16, for example, a NAND flash memory which is one of nonvolatile semiconductor memories is used. The NAND flash memory may be a single chip or may be composed of a plurality of chips in order to increase the storage capacity. For the first memory block 16, other than this, if it is a solid-state chip LSI having storage non-volatility, for example, MRAM (Magnetic Random Access Memory) and ReRAM (Resistive Random Access Memory) can be used. It is not limited to.

  The host interface 11 can accept a normal data operation request by address designation from the host system, that is, a data write and read request, and a write and read request to the key-value store type data in the metadata table 162. .

  Requests for writing to and reading from the first memory block 16 are received and controlled by the memory controller 13. Further, a fixed-length data generator 14 and a register memory 15 are connected between the memory controller 13 and the first memory block 16. The register memory 15 is also called a page register, R / W register, page cache, or the like, but is temporarily used as a storage area when writing or reading. In particular, the register memory 15 has an arithmetic function and is generally used for controlling the multi-value operation of the NAND flash memory, and is similarly used in the present embodiment.

  The present embodiment is characterized in that a fixed-length data generator 14, for example, a hash generator, used to write and read key-value type data to and from the first memory block 16 is provided. The hash generator functions as an address acquisition circuit that acquires an address at which key-value type data is stored in response to a key input. Here, the hash generator may be regarded as an electronic circuit having a function of generating a hash function, but a dedicated circuit may be used, or a general-purpose arithmetic circuit into which a hash function algorithm is input may be used. .

  The hash value (address) generated by the hash generator may collide. The memory controller 13 includes a comparison circuit and an address management circuit for processing after a hash value collision. A data storage / retrieval method using the fixed-length data generator (hash generator) 14 and the address management circuit will be described later. Although an example in which the memory controller 13 and the fixed-length data generator 14 are separately formed is shown here, the memory controller 13 may include the fixed-length data generator 14.

  Furthermore, as a configuration of the present embodiment, a local controller 12 is provided to control signal transmission / reception between the host interface 11 and the first memory block 16. The local controller 12 can include an error correction code decoding (ECC) circuit for data output from the first memory block 16. Note that the ECC circuit need not be provided in the local controller 12 as long as the ECC circuit is provided in the memory controller 13.

  In addition, the local controller 12 can have a function of managing a logical address-physical address conversion table that converts a logical address of the first memory block 16 into a physical address. Thereby, the local controller 12 can manage the correspondence between the actual data area 161 and the metadata table 162 and the logical address. That is, the local controller 12 distinguishes between the storage areas of the real data area 161 and the metadata table 162 in the logical address-physical address conversion table. For this reason, in the first memory block 16, the storage area of the actual data area 161 and the metadata table 162 does not need to be separated and may be mixed.

  For these processes, a second memory block can be included within the local controller 12. Alternatively, the second memory block may be connected to the outside of the local controller 12 via a bus line.

  The presence of the second memory block is obvious in the case of a conventional SSD, but even if the second memory block does not exist, it is not necessarily inconvenient in showing the minimum configuration of the present embodiment. However, the second memory block is not shown. However, if the local controller 12 can use the second memory block, the metadata table can be read from the first memory block 16 into the second memory block for reference.

  The second memory block is also used to compensate for a speed difference between the communication speed of the host interface 11 and the access speed of the first memory block 16. For this reason, the second memory block has a volatile / small capacity but a high access speed as compared with the first memory block 16.

  For example, a volatile DRAM or SRAM is used for the second memory block. Alternatively, if equivalent speed and capacity can be obtained, nonvolatile RAM (Random Access Memory), for example, MRAM (Magnetic Resistance Change Film RAM), ReRAM (Resistance Change Film RAM), FeRAM (Ferroelectric Film RAM), PCRAM (Phase change film RAM) or the like may be used. In the case of a memory system using a flash memory as the first memory block 16, it is common to provide a wear leveling function using the local controller 12, the second memory block, and a logical address-physical address conversion table. This may be used in the present embodiment.

<Data storage / retrieval method using hash function>
A data storage / retrieval method using the fixed-length data generator (hash generator) 14 and the address management circuit of this embodiment will be described. The address management circuit is provided in the memory controller 13 and performs processing for avoiding collision of hash values (addresses).

  Since this embodiment includes a hash generator, arbitrary length bit data can be converted into fixed length bit data by a hash function. Here, an example is shown in which the hash generator uses this function to generate a metadata address of fixed-length bit data from metadata of arbitrary-length bit data.

  As the hash function, a cryptographic hash function that is as uniform and sparse as possible is preferable. For example, SHA-1 (Secure Hash Algorithm-1), SHA-2 (Secure Hash Algorithm-2), MD4 (MessageDigest4), MD5 (MessageDigest5), etc. are used.

  The hash generator has a function that obtains a given <key> of arbitrary length as a hash (<key>) bit string of fixed length bits according to a hash function, and further shortens it to the desired bit length (BitLength). Have. For example, it has a division function of the following equation.

<key ID> = hash (<key>) mod BitLength
Alternatively, a portion of a desired length may be simply cut out from the beginning of the generated fixed-length bit string.

  If the length of the keyID generated in this way is made equal to the address length of the metadata table, it can be used as it is as the address of the metadata table. For example, in FIG. 1, the key-value type data in the metadata table will be described. If the result of generation of fixed-length bit data of “book” by the hash generator (hereinafter referred to as hash (“book”)) is 001. For example, address $ 001 in the metadata table corresponds to the key “book”. The corresponding value is stored at address $ 001.

  Similarly, in the case of “Blue”, if the result of hash (“Blue”) is 002, the key and value are stored in the manner that the value corresponding to the metadata address of $ 002 is saved.

  When searching for a key, for example, in the case of “book”, the output value (hash value) 001 of the hash function is the metadata address of the storage location as it is, so that address can be directly referred to. Such a data reference method utilizing the correspondence between the hash function and the memory address is called Hash-CAM.

  In Hash-CAM, even if a hash function that is as sparse as possible is used, the possibility that hash values (addresses) collide stochastically is not zero. In practice, the simplest and most effective way to reduce the possibility of hash value collisions is to provide a sufficiently large memory space. In practice, collisions can occur because the memory size is limited. In order to provide a post-collision processing function, this can be solved by providing a comparison circuit and an address management circuit having the following functions. When the hash values collide, the comparison circuit retrieves data by referring to the value, and compares and collates whether or not the retrieved data matches the key. The address management circuit changes the hash value (address) when the extracted data matches the key.

  For example, when “note” is stored as another key in addition to the above, the result of hash (“note”) also becomes 001. In this case, since $ 001 is already used for “book”, it is necessary to jump to another metadata address. Therefore, for example, the next metadata address is moved, that is, the address is incremented. In the example of FIG. 1, since $ 001 collides, it is checked whether $ 002 is free. However, since $ 002 is already used by hash (“Blue”), the next address $ 003 is further examined. As a result, if $ 003 is an empty address, the value corresponding to “note” is stored there.

  If this method is used, data can be stored even if hash values collide. However, it is necessary to devise a search for key-value type data. When searching for “note”, if hash (“note”) becomes 001, the metadata address of $ 001 is referred to, but the value of “book” that has already been stored is obtained by mistake.

  Therefore, it is necessary to check whether the correspondence between key and value is correct. Since the value of “book” is & 101 of the actual data address, data is read with reference to & 101. Since [book] is stored in the head data, it is determined that it is not a “note” key-value pair. Therefore, in order to find the key-value pair of “note”, the next metadata address $ 002 is also checked in the same manner, and it is found that it is not the key-value pair of “note”. As a result, it turns out that $ 003 of the next metadata address is a correct key-value pair. In this way, even when hash values collide, it is possible to search for key-value type data.

  It should be noted that [book], [Blue], [note] in the actual data part only need to be able to collate the key, so in effect, the first few bytes such as bo for book, Bl for Blue, no for note, etc. May be cut off and used in a fixed length. However, it should be noted that the possibility of collision is not zero when the length is fixed.

  In the above, the method of incrementing the address after the collision is used. However, as shown in a modification example to be described later, even if the key, that is, the address collides, if this embodiment is used, the actual data from the corresponding value is used. Since the address can be referred to, a method of storing the key itself in the actual data address and collating can also be used. Even in this case, it is necessary to collate the key as the search entry with the key in the actual data. An address management circuit as a Hash-CAM is also required. Therefore, although the system is slightly different, the same configuration as the Hash-CAM described above may be used.

  As described above, a hash value (address) is generated using a hash generator, and a hash value collision avoidance procedure is added to the storage of the metadata by the address management circuit, thereby efficiently in the memory system 10. A key-value store method can be realized.

  In this embodiment, in order to implement Hash-CAM, the memory controller 13 includes a hardware function (fixed length data generator) and a circuit that avoids collision of fixed length data (address). Here, the hardware function (fixed length data generator) may be provided in the memory controller 13, and the key-value type data at that time may be stored in the register memory 15 or directly. This may be performed on the first memory block 16.

  If the function of the key-value store method is not impaired in the above configuration, the temporary storage of the logical address-physical address conversion table and the wear leveling process are not limited to being performed inside the memory system 10, and the host system You may carry out using CPU and main memory. Further, a direct memory access controller (DMAC) may be provided in order for the memory system 10 to actively perform the key-value store method.

  In this embodiment, each functional block is connected by a bus line. Basically, it is desirable to have a fast and efficient bus line arrangement in the memory system. However, two or more types of bus lines are used in the memory system due to, for example, the difference between the chip interface standard and the external interface standard. May be.

  According to the present embodiment, the key-value type data for the metadata associated with the data makes it possible to perform the process of retrieving the data from the memory system simply and at high speed, and the user arbitrarily handles the metadata, It is possible to provide a memory system in which an accessible physical memory space can be used with maximum efficiency. That is, it is possible to provide a memory system in which an operation request for metadata is accepted by a memory system, and can be processed and output efficiently and at high speed by a key-value store system inside the memory system.

[Second Embodiment]
This embodiment is different from the first embodiment in hardware configuration, and includes a hardware CAM having a key verification dedicated memory space.

FIG. 5 is a block diagram illustrating a hardware configuration of the memory system according to the second embodiment.
The memory controller 13 controls mutual signal transmission / reception between the local controller 12 and the first memory block 16. In addition, a register memory 15 for writing / reading the first memory block 16 is provided. The search request can be temporarily stored in the register memory 15 and read matching can be determined. It is possible to perform coincidence determination by performing parallel reading in byte units and performing coincidence determination, and sequentially processing search data of arbitrary length.

  In the present embodiment, as shown in FIG. 5, a content associative memory (CAM) 24 is arranged at a location accessible from the memory controller 13. A content associative memory (CAM) is different from a normal memory that outputs data specified by an address in response to an input of an address, and matches the search data and all stored data in response to an input of the search data. This is a special-purpose memory for high-speed search that has the function of performing inconsistent comparison operations in parallel and outputting the address of the stored data that matches. Here, the CAM 24 functions as an address acquisition circuit that acquires an address in which key-value type data is stored in response to a key input. Further, the CAM can output the presence / absence of matching data as a “Match Flag” for data matching search. In the present embodiment, the CAM 24 is called a hardware CAM because it is realized by an electronic circuit for realizing these functions.

  The hardware CAM is directly connected to the register memory 15 and is disposed between the memory controller 13 and the register memory 15. Although an example in which the memory controller 13 and the CAM 24 are separately formed is shown here, the memory controller 13 may include the CAM 24.

  Further, since the first memory block 16 is a random access memory (RAM), the CAM 24 and the first memory block 16 function as a CAM-RAM. The CAM-RAM is a system in which an address is output by the CAM 24 and data is output by an addressed access RAM. The CAM address decoder and RAM address encoder are designed so that one CAM entry and one RAM entry have a one-to-one correspondence.

  In the present embodiment, the key is stored in the CAM 24, and a set of values corresponding to the key is stored in the first memory block 16, or the value (value) from the first memory block 16 is stored in the register memory 15. And function as a CAM-RAM.

  In order to use the hardware CAM, the metadata table needs to be transferred from the first memory block 16 to the register memory 15. If such a CAM-RAM is used, an address collision does not occur in principle as in the Hash-CAM in the first embodiment.

  Accordingly, since the key and value collation procedures and the search re-execution do not occur, the search becomes faster. Further, in the Hash-CAM, as a means for avoiding collision, there is often a margin in the metadata address. However, in the case of the hardware CAM, since there is no collision, the CAM 24 can be used efficiently.

  In this embodiment, since the CAM 24 is used only for key retrieval of key-value type data, the CAM is connected to a data input / output page register (register memory), so that the physical address space of the first memory block is obtained. Can be used to the maximum extent without partially occupying the key-value type data. Other configurations and effects are the same as those of the first embodiment described above.

[Third Embodiment]
As in the second embodiment, the present embodiment includes a hardware CAM having a key verification dedicated memory space, but the first memory block 16 includes a hardware CAM.

FIG. 6 is a block diagram illustrating a hardware configuration of the memory system according to the third embodiment.
As shown in the figure, the first memory block 16 has a real data area 161 and a CAM-RAM 163 for storing a metadata table. The CAM-RAM is used in the same manner as in the second embodiment. In this embodiment, a part of the first memory block 16 is a dedicated address space for CAM. Therefore, it is possible to search directly stored keys without moving the keys of the metadata table to the outside of the first memory block 16.

  In addition, it is possible to perform a complete parallel search by devising a method of giving collation data to the memory cell portion of the first memory block 16. For example, when the first memory block 16 is configured by a NAND flash memory, the read circuit is configured so that the area used for the CAM unit can be input as search data to all the gates simultaneously. . As a result, the output can be detected only for the NAND string for which the search is hit, and the CAM-RAM can be realized by associating the output with the page address of the RAM unit. Other configurations and effects are the same as those of the first embodiment described above.

[Fourth Embodiment]
This embodiment includes a fixed-length data generator (for example, a hash generator) as in the first embodiment, but the installation location of the fixed-length data generator is different from that of the first embodiment, and the local controller 12 A fixed-length data generator 14 is provided.

FIG. 7 is a block diagram illustrating a hardware configuration of the memory system according to the fourth embodiment.
As shown in the figure, the fixed length data generator 14 is arranged in the local controller 12. The fixed-length data generator 14 and the local controller 12 are not necessarily physically located in the same chip. It suffices if the local controller 12 is at a position where the fixed length data generator 14 can be easily accessed as compared with other functional blocks.

  The local controller 12 includes a buffer memory 121 as a second memory block. For this reason, the local controller 12 can store the logical address-physical address conversion table read from the first memory block 16 in the buffer memory 121 and perform logical address-physical address conversion. Similarly, wear leveling processing of the NAND flash can be performed. Furthermore, the local controller 12 can manage the correspondence between the metadata table 162 and logical addresses.

  As a feature of the present embodiment, since the fixed-length data generator 14 is in the local controller 12, at the time of logical address-physical address conversion, key hashing and association with value, that is, the metadata table 162 The creation of key-value type data can be efficiently performed in the local controller 12.

  In the present embodiment, two methods can be used: a method in which Hash-CAM is performed in the buffer memory 121 which is the second memory block, and a method in which the first memory block 16 is performed. Since the latter is the same as that of the first embodiment, the description is omitted, and the former will be described below.

  When creating the metadata table, data is read from the first memory block 16, stored in the buffer memory 121, and converted into a hash value. Since Hash-CAM is performed in the buffer memory 121, the metadata address corresponds to the physical address of the buffer memory 121.

  The created metadata table is written back to the first memory block 16 or stored in the buffer memory 121 as the second memory block. Thereby, key-value type data can be referred to the metadata table.

  When the metadata table is smaller than the size of the buffer memory 121, the key-value type data can be referred to in the buffer memory 121 that is faster than the first memory block 16, so that the search can be performed faster. Become.

  Further, when the buffer memory 121 is a non-volatile RAM, the power of the memory system can be turned off without writing the metadata back to the first memory block 16. Even after the power is turned on again, the metadata table is stored in the buffer memory 121, so that the process of reading from the first memory block 16 is not necessary. This results in an overall speed improvement. Other configurations and effects are the same as those of the first embodiment described above.

  In the present embodiment, the functions necessary for the Hash-CAM are provided near the local controller. However, the Hash-CAM is not necessarily performed on the buffer memory, and the first memory block is the same as in the first embodiment. It may be done against. When the metadata table is small, it is faster to perform Hash-CAM after reading all data into the buffer memory table. However, when the metadata table is larger than the buffer memory size, the Hash-CAM operation is directly performed on the first memory block. In some cases, it is faster.

[Fifth Embodiment]
In the present embodiment, as in the fourth embodiment, key-value type data is referred to by the buffer memory 121, which is the second memory block, but the local controller 12 includes a hardware CAM.

FIG. 8 is a block diagram illustrating a hardware configuration of the memory system according to the fifth embodiment.
As shown in the figure, a CAM 122 is connected to the buffer memory 121. In the figure, the buffer memory 121 and the CAM 122 are written separately, but they may be physically coupled.

  The output (hit signal) of the CAM 122 is directly connected to a part of the buffer memory 121 (for example, about half the memory capacity), and the CAM 122 and a part of the buffer memory 121 constitute a CAM-RAM. Therefore, it is possible to read data by specifying data (content).

  By using the hardware CAM 122, an address collision like Hash-CAM does not occur in principle. For this reason, since the key and value collation procedures and the search re-execution do not occur, the search becomes faster.

  Furthermore, while the buffer memory 121 has high access performance, the memory capacity is smaller than that of the first memory block 16, so that efficient operation of the memory space is required. In this embodiment, by adding the CAM 122, the memory space of the buffer memory 121 can be used with maximum efficiency. Other configurations and effects are the same as those of the second embodiment described above.

[Sixth Embodiment]
The present embodiment has substantially the same hardware configuration as the first embodiment, but differs in that there is no local controller in the memory system 10.

FIG. 9 is a block diagram illustrating a hardware configuration of the memory system according to the sixth embodiment.
As shown in the figure, the host interface 11 is directly connected to the memory controller 13.

  The method for realizing the key-value store method is the same as that in the first embodiment, but the handling of the logical address-physical address conversion table is different. Since the local controller and the second memory block do not exist in the memory system 10, the logical address-physical address conversion table is read from the first memory block 16 and is external to the memory system 10, for example, in the main memory 102. Handled by.

  Since Hash-CAM is performed using the fixed-length data generator 14 in the memory controller 13 as in the first embodiment, key-value type data is stored in the metadata table 162 within the memory system 10. Is called. Not only the key-value type data but also the changes in the metadata table 162 are returned to the host system, reflected in the logical address-physical address conversion table, and then written back to the first memory block 16 as appropriate.

  In the present embodiment, since the memory system 10 does not have a buffer memory or a local controller and has simplified functions, the memory system itself is compact.

  Note that a direct memory access controller (DMAC) is provided in order for the memory system 10 to actively execute the key-value store method, and the DMAC controls data transfer between the memory system 10 and the main memory 102. good. Other configurations and effects are the same as those of the first embodiment described above.

[Seventh Embodiment]
The present embodiment has substantially the same hardware configuration as the second embodiment, but differs in that there is no local controller in the memory system 10.

FIG. 10 is a block diagram illustrating a hardware configuration of the memory system according to the seventh embodiment.
The method for realizing the key-value store method by the hardware CAM is the same as that of the second embodiment, and the feature of the function due to the absence of the local controller is the same as that of the sixth embodiment.

[Eighth Embodiment]
The present embodiment has substantially the same hardware configuration as that of the third embodiment, but differs in that no local controller is present in the memory system 10.

FIG. 11 is a block diagram illustrating a hardware configuration of the memory system according to the eighth embodiment.
The method for realizing the key-value store method using the CAM-RAM is the same as that of the third embodiment, and the feature of the function due to the absence of the local controller is the same as that of the sixth embodiment.

  As described above, in the present embodiment, the key-value store scheme is realized by metadata, its table, a hash generator (Hash-CAM) as a searcher, and a hardware CAM (CAM-RAM).

  When the key-value store method is realized by using the first to eighth embodiments described above, the following modifications can be further taken, which will be specifically described below. In addition, the said embodiment is preferentially interpreted rather than a modification as the main point of this application.

[Modification 1]
FIG. 12 is a diagram schematically illustrating the relationship between the actual data area 161 in the first memory block 16 and the metadata table 162 and the mechanism of the key-value store method in the first modification.

  In the actual data area 161 shown in FIG. 12, a data file, a file name or file ID corresponding to the file, a key extracted from the file, and a metadata address storing the key are stored in the actual data address. ing. Such a storage method can be performed by an instruction from the host system, and can be realized if the memory system 10 has a key extraction function and a metadata address assignment function in advance.

  The metadata table 162 includes a key extracted when a file is stored at a real data address, a real data address where the key exists as a value corresponding to the key, and another key-value type data associated with the key. And the metadata address where is stored.

  With such address management, when the local controller 12 or the memory controller 13 instructs a key search request, the memory controller 13 searches for the key from the metadata address.

  For example, to obtain a file name including “book”, first, “book” is searched from the metadata address. “Book” is stored at the metadata address $ 002. From this metadata address $ 002, real values & 001 and & 002 and metadata addresses $ 011 and $ 012 are obtained as values.

  Here, a real address can be returned as a search result. Furthermore, by following the metadata address of the value, the set name to which “book” belongs can be traced. For example, at $ 011, a key “a-file.txt” and the actual data address and metadata address of the key can be obtained.

  Thus, the value (actual data) obtained as a search result can be obtained by continuously tracing the metadata table 162. In this modification, only the key exists in the metadata table, and the actual key is stored at the real address in the real data area 161.

[Modification 2]
FIG. 13 is a diagram schematically illustrating the relationship between the actual data area 161 and the metadata table 162 in the first memory block 16 and the mechanism of the key-value store method in the second modification.

  In the actual data area 161 shown in FIG. 13, a file and a file name or file ID corresponding to the file are stored in the actual data address. The metadata table 162 stores a key extracted from a file, a real address where a key corresponding to the key exists, and a metadata address for storing another set or element associated with the key.

  As in the first modification, when the local controller 12 or the memory controller 13 instructs a key search request, the memory controller 13 searches for the key from the metadata address.

  For example, to obtain a file name including “book”, first, “book” is searched from the metadata address. “Book” is stored at the metadata address $ 002. From this metadata address $ 002, & 011 of the actual data address and $ 011 and $ 012 of the metadata address are obtained as values.

  Here, the actual data address indicates not the storage location of the file to which “book” belongs, but the storage location in the actual data address of the key. On the other hand, the metadata address of value indicates a set to which “book” belongs. Therefore, the metadata address is traced and the actual data addresses of the values $ 011 and $ 012 are obtained and returned as search results, or the actual data addresses are traced and & 001 and & 002 data are returned. become. Modification 2 has an advantage that the amount of data per address in the real data area (real address space) 161 is smaller than that in Modification 1.

[Modification 3]
FIG. 14 is a diagram schematically showing the relationship between the actual data area 161 and the metadata table 162 in the first memory block 16 and the mechanism of the key-value store method in the third modification.

  This modification shows that the actual state of metadata is stored at the actual data address in the actual data area 161.

  In the actual data area 161 shown in FIG. 14, a file, a file name or file ID corresponding to the file, and a key metadata address are stored in the actual data address.

  In the metadata table 162, the metadata address stores the actual data address where the key extracted from the file and the corresponding key as a value exist. In addition, a physical address corresponding to the metadata address is shown.

  As described above, when the metadata address is assigned to a physical address space different from the actual data address, the correspondence table between the metadata address and the physical address is stored in the first memory block 16, and the memory system (local System) or the host system calls and uses the correspondence table.

  In addition, a key search method in the metadata table and an example in which values are stored at consecutive addresses are shown.

  The key-value type data in the metadata table 162 can also be realized as shown in FIG. The metadata address space is assumed to correspond to a specific physical address space of the first memory block 16. It is assumed that search target areas of continuous metadata addresses are divided by a unit called Slot. The key is stored in the head area in the slot.

  In the example shown in FIG. 15, the key “pen” is searched. The key is given by the memory controller 13 character by character, and the entire slot is searched. The key search mechanism in this example uses the hardware CAM in the memory controller 13 and the register memory 15 described in the embodiment.

  The slot where the first “p” of “pen” is obtained sets a match flag, and the slot searches for “e” next. If there is a continuous hit, the operation of setting the flag in the same manner and searching for the next is continued. Thus, in FIG. 15, a key match is obtained at # 003 and # 102, and value is obtained by continuously reading as it is with the slot. Since a control code is inserted in the separator between the key and the value, it can be distinguished.

  If this is expanded, a search with don't care mask bits can also be performed.

[Modification 4]
In this modification, key-value type data storing a fixed-length key and full-text search using the key-value type data are shown as an example.

  The need for fixed-length bits depends on the search scheme. The full text search system is roughly divided into two types: (1) sequential search and (2) index search. Furthermore, it can classify | categorize with the method of index creation (indexing), (a) Morphological analysis, (b) N-gram, (c) Suffix method is known.

  Among these, the morphological analysis is a method of extracting words existing in a dictionary prepared in advance. N-gram does not require a dictionary and can divide a word into N elements and search for an arbitrary character string. For example, if the set to be searched is S, the division method of S is uni-gram if it is one character, bi-gram if it is two characters, N-gram if it is N characters. For example, when S = innovation is decomposed, if it is a bi-gram, its divided elements (tokens) become at, in, io, nn, no, n, on, ov, ti, va. The suffix method handles arbitrary lengths, but is suitable for index file compression.

  In the actual data area 161 shown in FIG. 16, a file and a file name or file ID corresponding to the file are stored at the actual data address. In the metadata table 162, the key extracted from the file by the N-gram method is stored in the metadata address.

  In this example, the key of metadata is a bi-gram that is decomposed and extracted from “innovation” stored in the actual data address. The real address is indicated as a value corresponding to the key, and the appearance position (position) in the file when it is extracted from the file by bi-gram is stored in pairs. In bi-gram search, after searching for keys, sort by the number of occurrences of keys, compare position information in the file to confirm that they are continuous keywords, and obtain a desired set of search terms. .

  Here, since the key has a fixed length, it can be hashed as it is to form key-value type data. In the example of FIG. 16, “at” and “in” are respectively stored in the metadata address in that order. Actually, the N-gram token may be represented by a bit string using a code such as ASCII or UTF-6, and may be shortened to be a metadata address.

  Thus, the key-value store method according to the present embodiment is compatible with N-gram handling a fixed length and is suitable for high-speed indexing. Full-text search is fast but takes time to index. The indexing appropriately reads out a metadata table that is an element and a set, searches for an element that needs to be added / updated / deleted, and edits the set. For this reason, since file access occurs frequently, by using the memory system of the present embodiment, an efficient key-value store method in the memory system can be realized without imposing a load on the host system. It is also possible to perform indexing.

  In the present embodiment, as an example of the usefulness of the key-value store method in the memory system, the full-text search procedure has been frequently described. However, the present embodiment is not necessarily a technique specialized for full-text search.

  The present embodiment provides a key-value store scheme and its specific configuration in order to efficiently manage metadata when data is stored in a memory system.

  Although several embodiments of the present invention have been described, these embodiments are presented by way of example and are not intended to limit the scope of the invention. These novel embodiments can be implemented in various other forms, and various omissions, replacements, and changes can be made without departing from the scope of the invention. These embodiments and modifications thereof are included in the scope and gist of the invention, and are included in the invention described in the claims and the equivalents thereof.

  DESCRIPTION OF SYMBOLS 10 ... Memory system (or local system), 11 ... Host interface, 12 ... Local controller, 13 ... Memory controller (or chip controller), 14 ... Fixed length data generator, 15 ... Register memory (or cache memory), 16 ... 1st memory block, 24 ... content associative memory (CAM), 101 ... CPU, 102 ... main memory, 121 ... buffer memory, 122 ... CAM-RAM, 161 ... real data area (real address space), 162 ... metadata Table, 163... CAM-RAM.

Claims (12)

In a system for receiving a command from a host via an interface and transferring data to the host,
a non-volatile memory for storing a metadata table including a key-value type data that is a set of a key and a value corresponding to the key, and a data area;
A controller accessing the non-volatile memory;
With
The key-value type data is stored in the storage location specified by the first address of the metadata table,
In the value stored in the storage location designated by the first address, a second address indicating the storage location in the data area is stored,
The storage location designated by the second address stores a third address indicating the storage location where the value is stored, and the key.
The controller searches for whether the key exists in the metadata table when a command for ordering registration of a value associated with the key is input from the host.
When the key does not exist in the metadata table, determine whether the value associated with the key is stored in the data area;
When the value is not stored in the data area, the value is stored in a storage location specified by the fourth address of the data area;
When the value is stored in the data area, the key and the fourth address storing the value are registered in a storage location specified by the fifth address of the metadata table . In a system for receiving a command from a host via an interface and transferring data to the host,
a non-volatile memory for storing a metadata table including a key-value type data that is a set of a key and a value corresponding to the key, and a data area;
A controller accessing the non-volatile memory;
With
The key-value type data is stored in the storage location specified by the first address of the metadata table,
In the value stored in the storage location designated by the first address, a second address indicating the storage location in the data area is stored,
The storage location designated by the second address stores a third address indicating the storage location where the value is stored, and the key.
When the controller receives a command from the host to add a value to the value set associated with the key, the controller searches whether the key exists in the metadata table;
When the key exists in the metadata table, refer to the storage location specified by the fourth address of the data area stored in the value corresponding to the key;
A check is made to see if there is an empty storage location designated by the fourth address. If there is no empty, a pointer for jumping to the fifth address of the data area of another value is stored. A system for adding the value to a designated storage location . In a system for receiving a command from a host via an interface and transferring data to the host,
a non-volatile memory for storing a metadata table including a key-value type data that is a set of a key and a value corresponding to the key, and a data area;
A controller accessing the non-volatile memory;
With
The key-value type data is stored in the storage location specified by the first address of the metadata table,
In the value stored in the storage location designated by the first address, a second address indicating the storage location in the data area is stored,
The storage location designated by the second address stores a third address indicating the storage location where the value is stored, and the key.
When the controller receives a command from the host to output the value associated with the key, the controller searches whether the key exists in the metadata table;
When the key exists in the metadata table, refer to the storage location specified by the fourth address of the data area stored in the value corresponding to the key;
A system for reading a value stored in a storage location designated by the fourth address . In a system including a key-value store including a key-value type data which is a set of a key and a value corresponding to the key,
A non-volatile memory for storing a metadata table including the key-value type data and a data area;
A controller accessing the non-volatile memory;
With
The first address of the metadata table indicates a storage location where the first key-value type data is stored,
In the value corresponding to the key stored in the storage location designated by the first address, the second address indicating the storage location of the metadata table of the second key-value type data associated with the key, and A third address indicating the storage location of the data area is stored;
The third address is stored in a storage location designated by the second address,
A system in which the value and the key are stored in a storage location designated by the third address . In a system including a key-value store including a key-value type data which is a set of a key and a value corresponding to the key,
A non-volatile memory for storing a metadata table including the key-value type data and a data area;
A controller accessing the non-volatile memory;
With
The first address of the metadata table indicates a storage location where the first key-value type data is stored,
In the value corresponding to the key stored in the storage location designated by the first address, the second address indicating the storage location of the metadata table of the second key-value type data associated with the key, and A third address indicating the storage location of the data area is stored;
In the storage location designated by the second address, a fourth address indicating the storage location of the data area is stored,
The value is stored in the storage location designated by the fourth address,
A system in which the key is stored in a storage location designated by the third address . In a memory system including a key-value store including a key-value type data that is a set of a key and a value corresponding to the key,
A non-volatile memory for storing a metadata table including the key-value type data and a data area;
A controller accessing the non-volatile memory;
With
The first address of the metadata table indicates a storage location where the key-value type data is stored,
The value corresponding to the key stored in the storage location designated by the first address stores a second address indicating the storage location of the data area,
The storage location specified by the second address stores a third address indicating the storage location of the metadata table,
The storage location designated by the third address stores a fourth address indicating the storage location of the data area,
The value is stored in the storage location designated by the fourth address,
A system in which a physical address corresponding to the first address is stored in a storage location designated by the first address . In a memory system including a key-value store including a key-value type data that is a set of a key and a value corresponding to the key,
A non-volatile memory for storing a metadata table including the key-value type data and a data area;
A controller accessing the non-volatile memory;
With
The first address of the metadata table indicates a storage location where the key-value type data is stored,
An element obtained by dividing the key is stored as the key in a storage location designated by the first address,
A second address indicating a storage location of the data area is stored as a value corresponding to the key in a storage location designated by the first address;
The storage location specified by the second address stores the third address and the appearance position of the key in the value,
A system in which the value is stored in a storage location designated by the third address . The system according to claim 4, wherein the controller acquires the first address from the key using an address acquisition unit . The system according to claim 8, wherein the address acquisition unit is a hash generator that converts the key into the first address using a hash function . The address acquisition unit compares the key with the data stored in the address acquisition unit in response to the key input, and obtains the first address of the matched data by a content associative memory (CAM). The system of claim 8 . The system according to claim 1, further comprising a direct memory access controller that directly accesses the key-value type data without using the controller . The system according to claim 1, wherein the nonvolatile memory is a NAND flash memory .