KR20230011787A - Method and apparatus for switching migration policy - Google Patents

Abstract

Embodiments of the present invention provide a method and device for switching migration policy by considering application characteristics and memory writing characteristics in a hybrid memory system to which heterogeneous memories are applied. A migration policy switching device includes: a memory module including heterogeneous memories; and a memory manager which selects a migration policy suitable for the memory module based on stored application characteristics.

Description

Migration policy switching method and device {METHOD AND APPARATUS FOR SWITCHING MIGRATION POLICY}

The technical field to which the present invention belongs relates to a method and apparatus for switching a migration policy according to application characteristics and memory write characteristics.

The contents described in this part merely provide background information on the present embodiment and do not constitute prior art.

Volatile memory (VM), such as DRAM (dynamic random access memory) and DDR(N) (Double Data Rate SDRAM), has the advantage of high writing speed, but data is lost when power is not supplied, and data In order to maintain, there is a disadvantage in that power consumption increases because refresh must be periodically performed.

Non-Volatile Memory (NVM), such as Phase-Change Memory (PCM), has a higher density than volatile memory, so it is advantageous to implement high capacity, and data is maintained even when power is not applied, It has the advantage that it can be implemented as a low-power device because it does not consume standby power. However, NVM shows a similar level to VM in terms of read speed, but its write speed is much slower than that of VM, which is highly likely to cause bottlenecks. However, it has the disadvantage that it is not suitable for application to applications that occur frequently.

Considering the advantages and disadvantages of each type of memory, a hybrid method using different types of memory is applied. Hybrid memory includes a combination of VM and NVM and compensates for the disadvantages of VM and NVM. Hybrid memory uses the advantages of non-volatility, high integration, and low power consumption of NVM, but focuses on solving problems caused by slow write speed and limited number of writes by using VM.

In a hybrid memory system composed of various types of memory devices, it is necessary to change a migration policy according to the characteristics of an application.

Korean Patent Publication No. 10-2015-0043102 (2015.04.22) Korean Registered Patent Publication No. 10-1894132 (2018.08.27)

The main purpose of embodiments of the present invention is to switch a migration policy in consideration of application characteristics and memory write characteristics in a hybrid memory system to which heterogeneous memories are applied.

Other non-specified objects of the present invention may be additionally considered within the scope that can be easily inferred from the following detailed description and effects thereof.

According to an aspect of the present embodiment, a memory module including a heterogeneous memory; and a memory manager selecting a migration policy suitable for the memory module based on stored application characteristics.

The memory manager may include a migration engine performing migration according to the migration policy; and a remapping table for storing an original address and a new address of the data to be migrated.

The migration engine may include a control register for storing the application characteristics; a migration policy selector selecting one migration policy from among a plurality of migration policies based on the characteristics of the application; a migration policy storage unit that stores the plurality of migration policies; and a migration policy execution unit that proceeds with the selected one migration policy.

The application characteristics may include sensitivity, access frequency, write-intensity, locality, or a combination thereof.

The application characteristics may be stored in the control register by an operating system at the beginning of application execution.

When the migration policy execution unit proceeds with the selected one migration policy, the remapping table may be updated.

The memory module may include a volatile memory (VM) module including a plurality of volatile memory cells and a non-volatile memory (NVM) module including a plurality of non-volatile memory cells.

The migration policy execution unit classifies applications stored in the NVM module into predetermined segments, obtains and sums segment counts, which are the number of write commands included in each segment, and calculates an application count, which is a write count for the application; Selecting a segment having a segment count equal to or greater than a predetermined second threshold value among a plurality of segments of applications having the application count equal to or greater than a predetermined first threshold value, migrating the segment with data stored in the VM module, and remapping the data address of the migrated segment. The application stored in the NVM module is divided into predetermined segments, and segment counts, which are the number of write commands included in each segment, are obtained and summed to calculate an application count, which is a write count for the application, and the application count is A segment having a segment count equal to or greater than a predetermined second threshold may be selected from among a plurality of segments of applications having a predetermined first threshold or greater, migrated with data stored in the VM module, and a data address of the migrated segment may be remapped.

The VM module may include a VM controller controlling access to data stored in the VM module, and the NVM module may include an NVM controller controlling access to data stored in the NVM module.

The migration policy execution unit selectively migrates the segment of the application stored in the NVM module with the data stored in the VM module, and when an access command for the migrated data is applied, it can be modified with an access command for the remapped data address. .

The migration policy execution unit includes a command analysis unit that, when an access command including a read or write command is applied, analyzes the applied access command and divides it into a read or write command, application information, and write request information including data address. can do.

The migration policy executor receives the write request information, divides the authorized application into a plurality of segments, obtains a segment count and an application count, stores them in a write history table, and based on the stored light history table, the NVM module and a reassignment unit for selectively migrating segments of the stored program with corresponding data of the VM module and obtaining a data address that has been migrated and changed.

The migration policy executor remaps and stores the data address where segment data is stored and the changed data address, and when the read or write command is applied, the data address specified in the read or write command is converted to the remapped data address. It may include a remapping unit that is modified and outputs to the VM controller and the NVM controller.

The migration policy execution unit may include a swap buffer for receiving data migrated between the NVM module and the VM module, temporarily storing the data, and transferring the data to the counterpart module.

The reallocation unit analyzes the light history table, sets an application having an application count greater than or equal to the first threshold as a swap candidate application, and selects a segment having a segment count greater than or equal to the second threshold among a plurality of segments of the set swap candidate applications. It can be selected as the migration target segment from which migration is performed.

The reallocation unit transfers the data address of the migration target segment to the NVM controller, and the migration target segment stored in the NVM module is transferred to the VM module through the swap buffer and the VM controller under the control of the NVM controller. can be saved.

The reallocation unit checks the data addresses of the segments to be migrated, and if the data address difference between the sequentially migrated data is less than or equal to a predetermined reference address unit, the same data scope among the predetermined number of data scopes is accumulated and counted, and the reference If the address unit is exceeded, it is accumulated and counted again in other data scopes, and the address changes of sequentially delivered data are accumulated and recorded in a number of predetermined data scopes, the accumulated and recorded sizes in multiple data scopes are checked, and more than half of the data scopes You can adjust the size of the migration unit of the data to be migrated according to the checked interval by checking whether the cumulatively recorded size in .

When an access command is applied during migration, the migration policy executor may pause migration and perform an operation corresponding to the access command after completing transfer of data transferred to a swap buffer among data of a segment to be migrated.

The migration policy execution unit determines whether the data designated by the access command is stored in the swap buffer, and if it is determined that the data stored in the swap buffer is stored, accesses the data stored in the swap buffer and stores the data in the swap buffer. If it is determined that it is not, access to data stored in the VM module or the NVM module may be performed through the VM controller or the NVM controller.

According to another aspect of the present embodiment, in a migration policy switching method, acquiring application characteristics in a process of writing or compiling an application; inputting the application characteristics into a control register of a memory manager by an operating system at the beginning of application execution; and selecting a migration policy based on application characteristics stored in the control register.

The selecting of the migration policy may include obtaining a memory access count or an execution time; determining migration execution according to a result of comparing the number of times of access to the memory or execution time with a reference value; determining the migration candidate based on the characteristics of the application; performing migration according to the migration candidate; and updating the remapping table in which addresses of the data to be migrated are stored.

As described above, according to the embodiments of the present invention, in a hybrid memory system to which heterogeneous memories are applied, a migration policy can be switched in consideration of application characteristics and memory write characteristics.

Even if the effects are not explicitly mentioned here, the effects described in the following specification expected by the technical features of the present invention and their provisional effects are treated as described in the specification of the present invention.

1 and 2 are flowcharts illustrating a migration policy switching method according to an embodiment of the present invention.
3 to 6 are diagrams illustrating a migration policy switching device according to another embodiment of the present invention.
7 to 12 are diagrams illustrating a migration operation according to a migration policy considering write characteristics of a migration policy switching device according to another embodiment of the present invention.
13 is a diagram illustrating a migration execution unit of a migration policy switching device according to another embodiment of the present invention.
14 is a diagram illustrating a reallocation unit of a migration policy switching device according to another embodiment of the present invention.
15 is a diagram illustrating a concept of determining a migration data unit size of a migration policy switching device according to another embodiment of the present invention.

Hereinafter, in the description of the present invention, if it is determined that a related known function may unnecessarily obscure the subject matter of the present invention as an obvious matter to those skilled in the art, the detailed description thereof will be omitted, and some embodiments of the present invention will be described. It will be described in detail through exemplary drawings.

1 and 2 are flowcharts illustrating a migration policy switching method according to an embodiment of the present invention. The migration policy switching method may be performed by a migration policy switching device.

The migration policy switching method includes acquiring application characteristics in an application writing process or compiling process (S10), inputting the application characteristics to a control register of a memory manager by an operating system at the beginning of application execution (S20), and a control register. and selecting a migration policy based on application characteristics stored in (S30).

After the step of operating the application (S21), the step of selecting a migration policy (S30) is performed.

The step of selecting a migration policy (S30) is the step of acquiring the number of memory accesses or execution time (S31), and the result of comparing the number of memory accesses or execution time with a reference value (for example, a condition exceeding the reference value) determines migration execution. (S32), determining a migration candidate based on application characteristics (S33), performing migration according to the migration candidate (S34), and updating a remapping table in which addresses of data to be migrated are stored. Step S35 may be included.

When the step of updating the remapping table (S35) is completed, the step of performing a general memory operation (S36) is performed.

If the result of determining whether migration is performed (S32) indicates that migration is not necessary, a normal memory operation (S36) is performed.

3 to 6 are diagrams illustrating a migration policy switching device according to another embodiment of the present invention.

The migration policy switching device includes memory modules 200 and 300 including heterogeneous memories and a memory manager 100 that selects a migration policy suitable for the memory module based on stored application characteristics.

The memory manager 100 may include a remapping table 110 that stores original addresses and new addresses of data to be migrated, and a migration engine 120 that performs migration according to a migration policy.

The migration engine 120 includes a control register 121 that stores application characteristics, a migration policy selector 122 that selects one migration policy from among a plurality of migration policies based on the application characteristics, and a plurality of migration policies that stores the migration policies. It may include a migration policy storage unit 123 and a migration policy execution unit 124 that proceeds with one selected migration policy.

A control register (Control Register, 121) contains information necessary for decision in the migration policy selection unit (122). Application characteristics may be stored in control registers by the operating system at the beginning of application execution.

Application characteristics can be obtained through compiler analysis or programmer writing. Application characteristics may include sensitivity, access frequency, write-intensity, locality, or a combination thereof. Application features may be in the form of files, may exist in the headers of compiled binaries, or may exist in the preprocessor of #pragma compiler directives.

The control register 121 exists as a memory mapped I/O in the system memory map and has its own address. The operating system stores the application characteristics in the control register through the driver of the memory manager at the start of application execution.

The migration policy selector 122 selects a migration policy to be applied based on control register information.

The migration policy executor 124 performs migration based on the selected migration policy and updates the remapping table 110 . When the migration policy execution unit proceeds with one selected migration policy, a corresponding address may be updated according to migration in the remapping table. The remapping table 110 stores original addresses and new addresses of corresponding data when migration is performed by the migration policy execution unit 124 .

The memory module may include a volatile memory (VM) module including a plurality of volatile memory cells and a non-volatile memory (NVM) module including a plurality of non-volatile memory cells.

The migration policy execution unit divides the applications stored in the NVM module 300 into predetermined segments, obtains and sums the segment counts, which are the number of write commands included in each segment, and calculates the application count, which is the write count for the application. , Selecting a segment having a segment count equal to or greater than a predetermined second threshold among a plurality of segments of applications having an application count equal to or greater than a predetermined first threshold, migrating the segment with data stored in the VM module, and remapping the data address of the migrated segment. The applications stored in the NVM module are divided into predetermined segments, and segment counts, which are the number of write commands included in each segment, are obtained and summed to calculate the application count, which is the write count for the application. A segment having a segment count equal to or greater than a predetermined second threshold may be selected from among a plurality of segments of an application having a threshold value of 1 or greater, migrated with data stored in the VM module 200, and a data address of the migrated segment may be remapped.

The VM module 200 includes a VM controller 210 that controls access to data stored in the VM module, and the NVM module 300 includes an NVM controller 310 that controls access to data stored in the NVM module. can include

The migration policy executor selectively migrates the segment of the application stored in the NVM module with the data stored in the VM module, and when an access command for the migrated data is applied, it can be modified with an access command for the remapped data address.

When an access command including a read or write command is applied, the migration policy execution unit 124 analyzes the applied access command and divides the read or write command into a read or write command, application information, and write request information including data address. An analysis unit 320 may be included.

The migration policy executor 124 receives the write request information, divides the authorized applications into a plurality of segments, obtains a segment count and an application count, stores them in a light history table, and based on the stored light history table, the NVM module It may include a reassignment unit 132 that selectively migrates segments of the stored program with corresponding data of the VM module and obtains data addresses that have been migrated and changed.

The migration policy executor 124 remaps and stores the data address where data of the segment is stored and the changed data address, and when a read or write command is applied, the data address specified in the read or write command is remapped. It may include a remapping unit 133 that modifies data addresses and outputs them to the VM controller and the NVM controller.

The migration policy execution unit 124 may include a swap buffer 134 that receives data migrated between the NVM module and the VM module, temporarily stores it, and transfers it to the counterpart module.

The reallocation unit analyzes the light history table, sets an application having an application count greater than or equal to a first threshold as a swap candidate application, and migrates a segment having a segment count greater than or equal to a second threshold among a plurality of segments of the set swap candidate application. It can be selected as the target segment for migration.

The reallocation unit may transfer a data address of a migration target segment to the NVM controller so that the migration target segment stored in the NVM module is stored in the VM module through the swap buffer and the VM controller under the control of the NVM controller.

The reassignment unit checks the data addresses of the segments to be migrated, and if the data address difference between the sequentially migrated data is less than or equal to a predetermined reference address unit, the same data scope among the predetermined number of data scopes is accumulated and counted, and the reference address If the unit is exceeded, it is accumulated and counted again in other data scopes, and the address change of sequentially delivered data is accumulated and recorded in a number of predetermined data scopes, the accumulated and recorded size in multiple data scopes is checked, and It is possible to check whether the cumulatively recorded size corresponds to a pre-designated unit maintenance interval, unit reduction interval, or unit increase interval based on the currently set migration unit, and adjust the migration unit size of data to be migrated according to the identified interval.

When an access command is applied during migration, the migration policy executor may pause migration and perform an operation corresponding to the access command after completing transfer of data transferred to a swap buffer among data of a segment to be migrated.

The migration policy execution unit determines whether the data specified by the access command is stored in the swap buffer, and if it is determined that it is stored in the swap buffer, accesses the data stored in the swap buffer, and if it is determined that it is not stored in the swap buffer , data stored in the VM module or NVM module can be accessed through the VM controller or NVM controller.

7 to 12 are diagrams illustrating a migration operation according to a migration policy considering write characteristics of a migration policy switching device according to another embodiment of the present invention.

The memory manager 100 determines which module, among the VM module 200 and the NVM module 300, stores data corresponding to a read command applied from the outside, and if it is determined that the data is stored in the module 200, the VM controller ( 210), and when it is determined that the data is stored in the NVM module 300, the read command is transferred to the NVM controller 310. Then, the corresponding data is received from the VM controller 210 or the VM controller 210 and output to the outside.

The memory manager 100 transfers data to be stored and a write command to the VM controller 210 or the VM controller 210 according to a write command, so that the data is stored in the VM module 200 or the NVM module 300 . Data applied with the write command is basically stored in the NVM module 300, but if the data to be written has been previously migrated and stored from the NVM module 300 to the VM module 200, the VM module 200 It can be updated by overwriting the previously stored data with data to be written.

The memory manager 100 selects data to be migrated to the VM module 200 from among data stored in the NVM module 300, and transfers the selection result to the VM controller 210 and the NVM controller 310, and the NVM module ( 300) and the VM module 200 mutually swap the selected data and the corresponding data.

The memory manager 100 includes a swap buffer that temporarily stores data transferred from the VM controller 210 and the NVM controller 310 so that data migration can be performed between the VM module 200 and the NVM module 300. can do.

The VM controller 210 controls the VM module 200 according to the management of the memory manager 100 . The VM controller 210 searches for and acquires data stored in the VM module 200 in response to a read command applied through the memory manager 100 and transfers the obtained data to the memory manager 100 . In response to the write command, the applied data is stored in a predetermined memory address of the VM module 200 . In this case, the VM controller 210 may include a buffer 220 for temporarily storing data obtained from the VM module 200 or data transferred from the memory manager 100 and stored in the VM module 200 .

The VM controller 210 refreshes the data stored in the VM module 200 at a predetermined cycle so that the data stored in the VM module 200 is not lost, and is applied to the hybrid memory device under the management of the memory manager 100. Before power is cut off, data stored in the VM module 200 may be authorized and transferred to the NVM controller 310 through the swap buffer of the memory manager 100 .

The NVM controller 310 controls the NVM module 300 according to the management of the memory manager 100 . Similar to the VM controller 210, the NVM controller 310 searches for and obtains data stored in the NVM module 300 in response to a read command applied through the memory manager 100, and transmits the obtained data to the memory manager 100. ), and stores the applied data in a predetermined memory address of the NVM module 300 in response to a write command. The NVM controller 310 may also include the NVM buffer 320 to temporarily store data acquired from the NVM module 300 or data transferred from the memory manager 100 and stored in the NVM module 300.

Unlike the VM controller 210, the NVM controller 310 does not need to perform a refresh operation on data stored in the NVM module 300. The NVM controller 130 may receive data stored in the NVM module 300 selected by the memory manager 100 and transfer it to the VM controller 210 through a swap buffer.

The VM module 200 includes a plurality of VM cells to store data while power is applied. The VM module 200 receives, stores or refreshes data under the control of the VM controller 210 of the memory manager 100, and transfers the stored data to the VM controller 210.

The NVM module 300 includes a plurality of NVM cells and can always store data regardless of whether or not power is supplied. The NVM module 300 receives and stores data under the control of the NVM controller 310 of the memory manager 100 and transfers the stored data to the NVM controller 310 .

The memory manager 100 classifies and manages program (or application) data in segments. A segment represents a unit of data that is selected to be migrated. A plurality of programs (Program 0 to Program 3) composed of a plurality of segments (Segment 0 to Segment 3) may be stored in the NVM module 300, and the memory manager 100 may store a plurality of programs stored in the NVM module 300. Segments may be selected from each of the programs (Program 0 to Program 3) and migrated to the VM module 200.

In FIG. 7, in the NVM module 300, one segment ((segment 3), (segment 0)) in the 0th and 1st programs (Program 0, Program 1) and two segments in the 2nd program (Program 2), respectively. It is shown assuming that (segment 1, segment 2) is selected and migrated to the VM module 200.

The migration policy switching device reduces the number of NVM writes and improves the write speed by considering the write characteristics of VM and NVM.

The migration policy switching device classifies programs (applications) into segments, selects migration target segments based on the number of write commands in the segmented segments, performs migration, and performs migration based on the migration target segment's data address change. to improve write efficiency.

The number of write commands included in each segment of the programs (Program A, Program C) stored in the NVM module 300 is counted. At this time, when a program (Program A, Program C) to be written from the outside is applied, the memory manager 100 stores the applied program in the NVM module 300 while counting the number of write commands in units of each segment, and the program Once stored, the program count can be obtained by adding the counted segment counts. Here, the number of write commands counted for each segment is referred to as a segment count. In addition, an accumulated sum of segment counts for a plurality of segments included in each program data (Program A, Program C) is calculated as a program count.

Referring to FIG. 8 , in Program A, segment counts for four segments are (51, 20, 32, 24), and the program count is calculated as 127. Also, in the C program (Program C), the segment count for the four segments is (49, 38, 77, 47), and the program count is calculated as 211.

When data is changed in a program stored in the NVM module 300, that is, when a change occurs in a segment, the memory manager 100 may update a segment count according to the changed segment and reflect it to the program count.

FIG. 9 shows an initial state of the NVM module 300, and thus shows a state in which programs (Program A to Program C) are not substantially stored in the NVM module 300, and all program counts are 0.

10 is a state in which programs (Program A to Program C) are stored in the NVM module 300, and the memory manager 100 checks the number of writes for each of a plurality of segments for each program (Program A to Program C) and segments A count is obtained, and a program count is calculated based on the obtained segment count. Here, it is assumed that program counts for each program (Program A to Program C) are 127, 79, and 211, respectively.

When the program count is calculated, the memory manager 100 searches for a program having a program count equal to or greater than a predetermined first threshold value (eg, 100) among the calculated program counts. When programs (Program A, Program C) having a program count equal to or greater than the first threshold are searched for, the searched programs (Program A, Program C) are selected as swap candidate programs.

A segment count of each segment included in the selected swap candidate programs (Program A, Program C) is equal to or less than the first threshold and a value equal to or greater than a predetermined second threshold (eg, 50) is searched, and the segment count is When a segment having a value equal to or greater than the second threshold is searched for, the searched segment is selected as a migration target.

In FIG. 11, it is assumed that the segment count of each segment in swap candidate programs (Program A, Program C) is the same as in FIG. In the C program (Program C), the second segment with a segment count of 77 is selected as a migration target. As an example, a case in which one segment is selected as a migration target in each program (Program A, Program C) is shown, but the number of segments selected as a migration target in each program (Program A, Program C) is not limited. In the selected swap candidate programs (Program A, Program C), all segments having a segment count greater than or equal to the second threshold may be selected as migration target segments.

Even if all segments having a segment count equal to or greater than the second threshold exist in programs (Program B) that are not selected as swap candidate programs (Program A, Program C), the memory manager 100 detects segments of the program (Program B) that are not selected as swap candidate programs (Program A, Program C). Since the count is not considered, it is not selected as a target segment for migration.

When a segment to be migrated is selected, the memory manager 100 operates the VM controller 210 and the NVM controller 310 so that the segment selected in the NVM module 300 is migrated to the VM module 200, as shown in FIG. is controlled to migrate the segment to be migrated selected in the NVM module 300 to the VM module 200. The migration target segment selected by the NVM module 300 is transferred to the swap buffer of the memory manager 100 through the buffer 320 of the NVM controller 310, and the migration target segment transferred to the swap buffer is transferred to the VM controller 210. may be applied to the VM module 200 through the buffer 220 and stored.

The memory manager 100 may change the unit size of migrated data when migrating the selected migration target segment.

13 is a diagram illustrating a migration execution unit of a migration policy switching device according to another embodiment of the present invention.

The command analysis unit 131 analyzes commands such as read or write applied from an external device, and transfers read and write requests to the remapping unit 133 and transfers write request information to the reallocation unit 132 . Write request information may include a data address, program information, and the like.

Programs (applications) stored in the hybrid memory device may be executed by an external processor or the like. During execution, data is changed by a write command, and the changed data must be stored in the hybrid device again. When the data changed by the write command is applied, the command analyzer 131 transmits write request information including the address of the changed data, program information, and segment information to the reallocation unit 132 so that the previously stored data is the changed data. to be updated to The command analysis unit 131 transfers the write request information to the reallocation unit 132 by remapping information about segments migrated to the VM module 200 by program counts and segment counts in programs stored in the NVM module 300. It is transmitted to the unit 133 so that the remapping unit 133 can normally write and update data.

If the authorized program is a new program not stored in the previous VM module 200 or the NVM module 300, the reallocation unit 132 divides the authorized program into segments and counts the number of write commands included in each segment. to obtain a segment count, and calculate and store a program count based on the acquired segment count. The reallocation unit 132 sets locations where a plurality of segments of a program are to be stored in the VM module 200 or the NVM module 300 according to the calculated program count and segment count.

The reallocation unit 132 may allocate addresses so that segments of an applied program are basically stored in the NVM module 300 . The reallocation unit 132 selects a swap candidate program based on the stored program count and segment count, selects a migration target segment from the selected swap candidate program, and stores the selected migration target segment in the VM module 200. address can be assigned.

Addresses may be allocated so that the segments of the non-selected program and the segments other than the segment to be migrated in the swap candidate program are stored in the NVM module 300 . The reallocation unit 132 allocates and stores addresses for all segments of the new program to be stored in the NVM module 300, and then selects segments to be migrated from among the segments of the program stored in the NVM module 300 and selects segments to be migrated to the VM module ( 200) may be migrated.

After the VM controller 210 or the NVM controller 310 stores the segment of the program to which the application has been applied, the reallocation unit 132 may receive and acquire the stored data address from the VM controller 210 or the NVM controller 310. .

The reallocation unit 132 analyzes program counts and segment counts of programs previously stored in the NVM module 300 to selectively migrate segments of programs stored in the NVM module 300 to the VM module 300. . When a new program is written, the reallocation unit 132 obtains and stores the program count and segment count in advance, and when a change in program segments is confirmed from the write request information, the program count and segment count are recalculated and updated. and save it The light history for each program segment is stored in advance. According to the stored write history, migration target segments are selected from among program segments and migrated to the VM module 300 . At this time, the reallocation unit 132 transfers the data address of the segment to be migrated from the NVM module 300 and the data address of the segment migrated to the VM module 200 to the remapping unit 133 so that the remapping unit 133 is migrated. Save the change of data address.

The reallocation unit 132 may analyze the data address of the migration target segment and adjust the unit size of migrated data. A detailed description of how the reallocation unit 132 adjusts the unit size of migrated data will be described later.

The remapping unit 133 receives and stores the data address changed by the migration from the reallocation unit 132, and when a read command or a write command is applied from the command analysis unit 131, the read command and the write command are based on the stored data address. The command is modified and transmitted to the VM controller 210 or the NVM controller 310 .

When a data address for the data migrated by the reallocation unit 132 is applied, the remapping unit 133 stores the existing data address and the migrated data address in the remapping table. Then, when a read or write command is applied from the command analyzer 131, it is determined whether a memory address reallocated for the corresponding data exists in the remapping table, and the memory address reallocated by the reallocation unit 132 exists. , the read or write command is modified into a command for the reallocated memory address and transmitted to the VM controller 210 or the NVM controller 310, so that the VM controller 120 or the NVM controller 310 changes the reallocated memory address. Allows data to be read or written in.

When the swap buffer 134 migrates data stored in the VM module 200 to the NVM module 300 or migrates data stored in the NVM module 300 to the VM module 200, the VM controller 210 and the NVM Data transmitted from the buffers 220 and 320 of the controller 310 are temporarily stored and then transferred to the counterpart buffers 220 and 320 to prevent data collision during migration.

14 is a diagram illustrating a reallocation unit of a migration policy switching device according to another embodiment of the present invention.

The reallocation unit 132 may include a write request information analysis unit 141 , a write history table 142 , a migration determination unit 143 and a migration request unit 144 .

The write request information analyzer 141 analyzes write request information applied from the command analyzer 131 . The write request information analyzer 141 checks program information, segment information, data addresses, etc. from the write request information and transmits them to the write history table 142 .

The write history table 142 classifies programs into segments from program information and segment information checked by the write request information analyzer 141, counts the number of write commands included in each segment, and obtains a segment count. , calculates a program count based on the acquired segment count. Then, the calculated program count and segment count are stored in the write history table 142 .

The migration determination unit 143 sets a swap candidate program from the program count stored in the write history table 142, and selects a segment to be migrated from the segment count of segments included in the set swap candidate program.

The migration request unit 144 transfers the data address of the migration target segment to the VM controller 210 and the NVM controller 310 so that the migration target segment is migrated to the VM module 200 or the NVM module 300 .

The migration request unit 144 may receive data addresses after migration from the VM controller 210 and the NVM controller 130 , store them in the write history table 142 , and transfer them to the remapping unit 133 . At this time, the migration request unit 144 may adjust the unit size of migrated data based on the data address of the data included in the segment.

15 is a diagram illustrating a concept of determining a migration data unit size of a migration policy switching device according to another embodiment of the present invention.

The migration request unit 144 may include a plurality of data scopes (Scope0 to Scope5). The number of data scopes (Scope0 ~ Scope5) can be adjusted in various ways. The data scope is a kind of virtual counter to determine whether data is sequentially applied, and each area distinguished from each of multiple data scopes (Scope0 ~ Scope5) is a count to set the data migration unit size that can be set when performing migration. indicates

The total size of each of the plurality of data scopes (Scope0 to Scope5) may be set to indicate the maximum data migration unit size (for example, 128Kbyte) that can be migrated. For example, in FIG. 9 , each area of each data scope (Scope0 to Scope5) is set to indicate a minimum data migration unit size of 4Kbye, so each of a plurality of data scopes (Scope0 to Scope5) can be divided into 32 areas. Since the migration unit size generally changes in the order of a power of 2, it may be set so that each area represents a migration unit size that increases by a power of 2 from 4Kbye. Each area of the data scope may be set to sequentially indicate the order of 4, 8, 16, 32, ..., 128 Kbytes.

The migration request unit 144 may set a first reference migration unit having a smaller value and a second reference migration unit having a larger value based on the currently set migration unit. That is, the section between the first reference migration unit and the second reference migration unit centered on the current migration unit is set as the unit maintenance section, the section less than the first reference migration unit is set as the unit reduction section, and the section equal to or greater than the first reference migration unit is set. Set the interval to a unit increment interval.

Thereafter, the migration request unit 144 sequentially fills each area in one data scope among a plurality of data scopes (Scope0 to Scope5) if the address difference between sequentially accessed data addresses is less than or equal to the currently set predetermined reference address unit. . If the data address difference continues to be less than or equal to the reference address unit, areas are continuously filled in the same data scope, whereas if the data address difference exceeds the reference address unit, each area is sequentially filled again in the next data scope.

The migration request unit 144 repeats the above process, and when each area is filled for all data scopes (Scope0 to Scope5), it can cycle again to sequentially fill the data scope areas (Scope0 to Scope5).

Migration request unit 144 calculates the size of the area filled in a plurality of data scopes (Scope0 to Scope5) during a predetermined period, and more than half of the plurality of data scopes (Scope0 to Scope5), that is, more than half of the data scopes are filled in The size of the region determines the included interval among the unit maintenance interval, the unit reduction interval, and the unit increase interval. If it is determined to be included in the unit maintenance section, the currently set migration unit is maintained, while if it is determined to be included in the unit reduction section or the unit increase section, the set migration unit is increased or decreased by a predetermined unit.

When the migration request unit 144 changes the migration unit, the size of data migrated at one time increases or decreases corresponding to the size of continuous data in the segment to be migrated, thereby reducing the number of migrations itself and the VM module (200 ) and the number of writes in the NVM module 300 to improve not only the write speed but also the durability of the NVM module 300 .

In the hybrid memory device management method, first, it is determined whether migration is to be performed on data of a currently stored program. Migration may be performed to mutually swap data stored in the NVM module 300 and data stored in the VM module 200 . When data stored in the NVM module 300 is migrated with the VM module 200, after the program stored in the NVM module 300 is read, the read program can be executed. In addition, migration may be performed even when a new program is applied to and stored in the hybrid memory device by a write command. And, if it is determined that migration is to be performed, the hybrid memory device executes migration in a predetermined manner.

In the migration step, first, the number of write commands in each of a plurality of segments constituting a plurality of programs is counted to obtain a segment count, and a program count for each program is calculated by summing the obtained segment counts.

A program whose calculated program count is equal to or greater than a predetermined first threshold is determined. If a program having a program count equal to or greater than the first threshold value is identified, the identified program is set as a swap candidate program. Thereafter, among a plurality of segments of the set swap candidate program, segments having a segment count greater than or equal to a second threshold value are determined. When a segment having a segment count equal to or greater than the second threshold value is determined, the determined segment is selected as a migration target segment.

Start migration for the migration target segment. When migration starts, address differences in data transmitted sequentially in segments are checked. If the checked address differences are less than or equal to a predetermined reference address unit, they are accumulated in the same data scope, and if they exceed the reference address unit, they are accumulated in another data scope. In this way, address changes of sequentially transmitted data are accumulated and recorded in a predetermined number of data scopes. In addition, the cumulatively recorded size of multiple data scopes is checked for each predetermined interval, and the cumulatively recorded size of more than half of the multiple data scopes is determined based on the currently set migration unit. Change the migration unit by checking which section of the unit increase section it corresponds to.

Determines whether a data access command such as a read or write command is applied. If it is determined that the access command is authorized, it is determined whether data migration is currently being performed between the VM module 200 and the NVM module 300. If it is determined that the migration is being performed, it waits until the data currently being migrated is stored in the swap buffer 134, and then the migration is paused.

It is determined whether data to which an access command has been applied is stored in the swap buffer 134. When it is determined that the data is stored in the swap buffer 134, access is performed to the swap buffer 134. That is, data is read or written. And the migration continues.

If it is determined that the data is not stored in the swap buffer 134, the data stored in the VM module 200 or the NVM module 300 is accessed through the VM controller 120 or the NVM controller 310. Also, even when migration is not currently being performed, data stored in the VM module 200 or the NVM module 300 is accessed through the VM controller 210 or the NVM controller 310 .

The migration policy switching device may be implemented in a logic circuit by hardware, firmware, software, or a combination thereof, or may be implemented using a general-purpose or special-purpose computer. The device may be implemented using a hardwired device, a field programmable gate array (FPGA), an application specific integrated circuit (ASIC), or the like. Also, the device may be implemented as a System on Chip (SoC) including one or more processors and controllers.

The migration policy switching device may be installed in a computing device or server equipped with hardware elements in the form of software, hardware, or a combination thereof. A computing device or server includes all or part of a communication device such as a communication modem for communicating with various devices or wired/wireless communication networks, a memory for storing data for executing a program, and a microprocessor for executing calculations and commands by executing a program. It can mean a variety of devices, including

In Figures 1 and 2, it is described that each process is sequentially executed, but this is merely an example, and a person skilled in the art will refer to Figures 1 and 2 without departing from the essential characteristics of the embodiment of the present invention. Various modifications and variations may be applied by changing the order described, executing one or more processes in parallel, or adding another process.

Operations according to the present embodiments may be implemented in the form of program instructions that can be executed through various computer means and recorded in a computer readable medium. Computer readable medium refers to any medium that participates in providing instructions to a processor for execution. A computer readable medium may include program instructions, data files, data structures, or combinations thereof. For example, there may be a magnetic medium, an optical recording medium, a memory, and the like. The computer program may be distributed over networked computer systems so that computer readable codes are stored and executed in a distributed manner. Functional programs, codes, and code segments for implementing this embodiment may be easily inferred by programmers in the art to which this embodiment belongs.

These embodiments are for explaining the technical idea of this embodiment, and the scope of the technical idea of this embodiment is not limited by these embodiments. The scope of protection of this embodiment should be interpreted according to the claims below, and all technical ideas within the scope equivalent thereto should be construed as being included in the scope of rights of this embodiment.

Claims (16)

a memory module including heterogeneous memories; and
A migration policy switching device including a memory manager that selects a migration policy suitable for the memory module based on stored application characteristics. According to claim 1,
The memory manager,
a migration engine that performs migration according to the migration policy; and
and a remapping table storing an original address and a new address of the data to be migrated. According to claim 2,
The migration engine,
a control register for storing the application characteristics;
a migration policy selector selecting one migration policy from among a plurality of migration policies based on the characteristics of the application;
a migration policy storage unit that stores the plurality of migration policies; and
and a migration policy execution unit that proceeds with the selected one migration policy. According to claim 3,
The application characteristics include sensitivity, access frequency, write-intensity, locality, or a combination thereof. According to claim 4,
The migration policy switching device, characterized in that the application characteristics are stored in the control register by the operating system at the beginning of application execution. According to claim 3,
The migration policy switching device of claim 1 , wherein the remapping table is updated when the migration policy execution unit proceeds with the selected one migration policy. According to claim 3,
The memory module includes a VM (Volatile Memory) module including a plurality of volatile memory cells and a Non-Volatile Memory (NVM) module including a plurality of non-volatile memory cells,
The migration policy execution unit classifies applications stored in the NVM module into predetermined segments, obtains and sums segment counts, which are the number of write commands included in each segment, and calculates an application count, which is a write count for the application; Selecting a segment having a segment count equal to or greater than a predetermined second threshold value among a plurality of segments of applications having the application count equal to or greater than a predetermined first threshold value, migrating the segment with data stored in the VM module, and remapping the data address of the migrated segment. The application stored in the NVM module is divided into predetermined segments, and segment counts, which are the number of write commands included in each segment, are obtained and summed to calculate an application count, which is a write count for the application, and the application count is Selecting a segment whose segment count is equal to or greater than a predetermined second threshold among a plurality of segments of applications having a predetermined first threshold or higher is migrated with data stored in the VM module, and data addresses of the migrated segments are remapped. A migration policy switching device that According to claim 7,
The VM module includes a VM controller controlling access to data stored in the VM module;
The NVM module includes an NVM controller that controls access to data stored in the NVM module;
The migration policy execution unit selectively migrates the application segment stored in the NVM module with the data stored in the VM module, and when an access command for the migrated data is applied, modifying the remapped data address with an access command. A characterized migration policy switching device. According to claim 8,
The migration policy execution unit,
When an access command including a read or write command is applied, a command analyzer analyzes the applied access command and divides the read or write command into write request information including application information and data address;
receiving the write request information, classifying the authorized application into a plurality of segments, obtaining a segment count and an application count and storing them in a light history table, and selectively selecting segments of programs stored in the NVM module based on the stored light history table a reassignment unit for migrating corresponding data of the VM module and obtaining a data address that has been migrated and changed;
The data address at which segment data is stored and the changed data address are remapped and stored, and when the read or write command is applied, the data address specified in the read or write command is modified to the remapped data address and the VM controller and a remapping unit outputting an output to the NVM controller; and a swap buffer for receiving data migrated between the NVM module and the VM module, temporarily storing the data, and transferring the data to the counterpart module. According to claim 9,
The reallocation unit analyzes the light history table, sets an application having an application count greater than or equal to the first threshold as a swap candidate application, and selects a segment having a segment count greater than or equal to the second threshold among a plurality of segments of the set swap candidate applications. A migration policy switching device characterized in that it selects as a migration target segment for which migration is to be performed. According to claim 10,
The reallocation unit transfers the data address of the migration target segment to the NVM controller, and the migration target segment stored in the NVM module is transferred to the VM module through the swap buffer and the VM controller under the control of the NVM controller. Migration policy switching device, characterized in that to cause to be stored. According to claim 11,
The reallocation unit checks the data addresses of the segments to be migrated, and if the data address difference between the sequentially migrated data is less than or equal to a predetermined reference address unit, the same data scope among the predetermined number of data scopes is accumulated and counted, and the reference If the address unit is exceeded, it is accumulated and counted again in other data scopes, and the address changes of sequentially delivered data are accumulated and recorded in a number of predetermined data scopes, the accumulated and recorded sizes in multiple data scopes are checked, and more than half of the data scopes It checks whether the cumulatively recorded size in the currently set migration unit corresponds to a preset unit maintenance interval, unit reduction interval, or unit increase interval, and adjusts the migration unit size of the data to be migrated according to the checked interval. A characterized migration policy switching device. According to claim 9,
The migration policy execution unit, when an access command is applied during migration, pauses migration after completion of transfer of data transferred to a swap buffer among data of a segment to be migrated, and performs an operation corresponding to the access command. Migration Policy Switching Device. According to claim 13,
The migration policy execution unit determines whether the data designated by the access command is stored in the swap buffer, and if it is determined that the data stored in the swap buffer is stored, accesses the data stored in the swap buffer and stores the data in the swap buffer. If it is determined that it is not, the migration policy switching device accesses data stored in the VM module or the NVM module through the VM controller or the NVM controller. In the migration policy switching method,
Acquiring application characteristics in a process of writing or compiling an application;
inputting the application characteristics into a control register of a memory manager by an operating system at the beginning of application execution; and
and selecting a migration policy based on application characteristics stored in the control register. According to claim 15,
In the step of selecting the migration policy,
obtaining the number of memory accesses or execution time;
determining migration execution according to a result of comparing the number of times of access to the memory or execution time with a reference value;
determining the migration candidate based on the characteristics of the application;
performing migration according to the migration candidate; and
and updating the remapping table in which addresses of the data to be migrated are stored.

Images