KR102565873B1 - Method for allocating memory bus connected storage in numa system - Google Patents

Abstract

A method of operating a computing device operated by at least one processor, comprising: managing a free storage space of each node in which a storage device is installed; requesting a file write from a process executed on an arbitrary node; and storage space according to the request for writing the file. Receiving an allocation request, and allocating space in a storage device belonging to the node in which the process is executed, or allocating space in a storage device belonging to a node different from the node in accordance with a preset storage space allocation policy. Including allocating

Description

A method for allocating a storage device used by connecting to a memory bus in a NUMA system

The present invention relates to a technology for allocating a storage device used by connecting to a memory bus in a NUMA system.

Applications of High Performance Computing use techniques such as checkpointing and logging to record the operating state of an application or system in a permanent storage device in preparation for failures that may occur unexpectedly. do. The use of such a failure response technique increases the stability of file use, but degrades overall file processing performance due to redundant writing. Therefore, fast file writing performance is required to minimize performance loss.

The use of fast storage devices is increasing for fast file writing. As an example, a many-core NUMA (Non-Uniform Memory Access) structure in which a memory storage device (Memory Bus connected Storage, hereinafter referred to as 'MBS') connected to a memory bus has dozens or more cores. installed on the system and used.

However, in a NUMA system with Linux installed, several stages of software processing path developed on the premise of using a slow storage device are still used, and high file processing performance is not obtained because MBS is used without recognizing that it is installed.

On the other hand, NUMA systems have different access latency depending on the distance between the central processing unit (CPU) and the device connected to the memory bus, and the access latency can affect the file processing performance of applications using MBS.

Therefore, there is a need to utilize the performance characteristics of a fast MBS in consideration of the characteristics of a NUMA system.

The problem to be solved is to provide a method for allocating storage space by selecting an MBS that has low access latency and high file processing performance in response to an MBS allocation request.

In addition, a problem to be solved is to provide a method for simplifying an inappropriate file processing path in consideration of the characteristics of a high-speed MBS.

A method of operating a computing device operated by at least one processor according to an embodiment, comprising: managing free storage space of each node in which a storage device is installed; requesting a file write from a process executed in an arbitrary node and writing the file; Receiving a storage space allocation request according to the request, and allocating space in a storage device belonging to the arbitrary node where the process was executed or belonging to a node different from the arbitrary node according to a preset storage space allocation policy Allocating space on the storage device.

In the managing step, each node may be divided into a candidate node having a storage space larger than a preset reference value and an obese node having a storage space smaller than or equal to the reference value.

The managing may include collecting installation information including a physical address of the storage device connected to a memory bus, and determining a node where the storage device is installed based on the installation information.

In the collecting step, the physical address at which the storage device is installed may be collected using the physical address and memory affinity information of the system.

The computing device operates as a system with a non-uniform memory access (NUMA) structure, and the storage device may be a memory bus connected storage (MBS) connected to a memory bus.

The storage space allocation policy is any one of a local policy for allocating space on a storage device belonging to the arbitrary node and a striping policy for sequentially allocating space on a storage device belonging to a plurality of nodes including the arbitrary node. can be one

The storage space allocation policy may be set to either the local policy or the striping policy in a process of executing an application through an I/O control interface of the computing device.

A method of operating a computing device operated by at least one processor according to another embodiment, comprising the steps of managing a free storage space of each node including a storage device and a central processing unit, a storage space from a process executed in an arbitrary node. Receiving an allocation request, and allocating a space in a storage device belonging to a node with the shortest access time from the arbitrary node where the process is executed, wherein the computing device performs non-uniform memory access (Non-Uniform Memory Access, NUMA) structured system.

The managing step may include monitoring the free storage space of each node, setting a node having a free storage space larger than a preset reference value as a candidate node, and selecting a node having a free storage space smaller than or equal to the reference value. It can be set as an obese node.

The storage device is a memory bus connected storage (MBS) used by connecting to a memory bus, and the central processing unit may include a plurality of cores.

The allocating may include checking the free storage space of the storage device belonging to the arbitrary node, and if there is no free storage space of the storage device belonging to the arbitrary node, the access time among the random node and other nodes It may include sequentially allocating the storage space of these short nodes.

In the allocating, a buddy memory allocation method of repeatedly dividing the free storage space by 2 and allocating a memory having a size closest to the size of the requested storage space may be used.

According to the present invention, since a file processing path is simplified in consideration of MBS characteristics in the Linux kernel, performance loss due to unnecessary overhead can be prevented and file processing performance can be improved.

In addition, according to the present invention, since storage space can be allocated from the MBS of a node capable of minimizing an access delay time, system performance can be improved.

1 is an explanatory diagram showing the structure of a NUMA system.
2 is an explanatory diagram of a method of using an existing MBS.
3 is a configuration diagram showing a software layer of an existing file system.
4 is a configuration diagram illustrating software layers of an environment in which a storage device management unit is installed according to an exemplary embodiment.
5 is a configuration diagram of a storage device management unit according to an exemplary embodiment.
6 is an exemplary diagram illustrating a structure in which an MBS is installed in a NUMA system according to an embodiment.
7 is an explanatory diagram of an operating method of an address management unit according to an exemplary embodiment.
8 is an explanatory diagram of a method for managing a node by a space management unit according to an embodiment.
9 is a flowchart of a method for managing a node by a monitoring unit according to an embodiment.
10 is an explanatory diagram of a method of allocating an MBS by a space management unit according to an embodiment.
11 is an explanatory diagram of a method of allocating an MBS by a space management unit according to another embodiment.
12 is a flowchart of a method of managing a node by a storage manager according to an embodiment.
13 is a flowchart of a method of allocating a storage space by a storage device manager according to an exemplary embodiment.
14 is a hardware configuration diagram of a computing device according to an embodiment.

Hereinafter, with reference to the accompanying drawings, embodiments of the present invention will be described in detail so that those skilled in the art can easily carry out the present invention. However, the present invention may be embodied in many different forms and is not limited to the embodiments described herein. And in order to clearly explain the present invention in the drawings, parts irrelevant to the description are omitted, and similar reference numerals are attached to similar parts throughout the specification.

Throughout the specification, when a certain component is said to "include", it means that it may further include other components without excluding other components unless otherwise stated. In addition, terms such as “… unit”, “… unit”, and “module” described in the specification mean a unit that processes at least one function or operation, which may be implemented as hardware or software or a combination of hardware and software. there is.

1 is an explanatory diagram showing the structure of a NUMA system.

Referring to (a) of FIG. 1, in a system with a Many-Core Non-Uniform Memory Access (hereinafter referred to as 'NUMA') structure having dozens or more cores, a memory bus A storage device that is connected and used (Memory Bus connected Storage, hereinafter referred to as 'MBS') may be installed. This system structure can be used in various fields requiring fast file data input/output processing.

At this time, it is divided into local access and remote access according to the location of the core and memory to be used. For example, access from the CPU on node 0 to DRAM connected to the memory bus of node 0 is local access, and access from the CPU on node 1 to DRAM connected to the memory bus on node 0 is remote access.

Referring to (b) of FIG. 1, when MBS is used in a NUMA structured system, it can be seen that access latency is different depending on the distance between the central processing unit (CPU) and the memory node. there is.

For example, the access time from node 0, node 1, node 2, and node 3 to the DRAM connected to the memory bus of node 0 is shortest at node 0 and longest at node 2, respectively.

Meanwhile, two methods are used to use MBSs installed in each node as a block storage device. It will be described with reference to FIG. 2 below.

2 is an explanatory diagram of a method of using an existing MBS.

Referring to (a) of FIG. 2, a file system designed on the premise of using a block storage device can use MBSs installed in each node in one address space using a logical volume manager.

If you mount the MBS installed on each node as a device using Logical Volume Manager, you can use all the storage space of the MBS installed in the system.

However, overhead occurs during the execution of the logical volume manager, and since this method is designed based on the use of traditional block storage devices, the MBS installed on the node where the application is running cannot be used. Therefore, there is a problem in that it is not possible to select and use an MBS having a short access delay time.

Referring to (b) of FIG. 2 , a user or an administrator may directly designate file locations of MBSs installed in the system for each node and use the MBSs installed in each node in different address spaces.

If the user or administrator recognizes that it is a NUMA system and sets the process to run only on the node where MBS is installed, the MBS with the lowest access latency can be used and high file processing performance can be obtained.

However, if there is no separate setting, the operating system cannot allocate the MBS space by itself considering the access delay between the MBS and the process in the NUMA structure. Therefore, it may not be guaranteed that the node where the process is running can use the MBS with the lowest access latency.

Therefore, in this specification, a method of using MBS is proposed in consideration of a NUMA environment.

3 is a configuration diagram illustrating a software layer of an existing file system, and FIG. 4 is a configuration diagram illustrating a software layer of an environment in which a storage management unit according to an embodiment is installed.

Referring to FIG. 3 , a motor-based slow storage device that has been used for a long time has an access delay time greater than that of dynamic random access memory (DRAM), which is a volatile memory, and is used by being connected to a device bus that is slower. In order to increase file processing performance, access to the storage device is minimized, and several levels of software layers exist for this purpose.

As an example, a logical volume manager was used to use multiple block-based physical storage devices in one logical address space.

The logical volume manager makes the physical devices used into volume groups and makes the volume groups into logical volumes in order to make the distributed space installed on each node into one logical address space. Overhead occurs in this process, which can act as a factor that degrades the overall file processing performance.

Referring to FIG. 4 , unlike FIG. 3 , there is no software layer such as a logical volume manager for using MBSs distributed to each node in one address space, and instead a storage device management unit 1000 exists.

The storage management unit 1000 performs the operations described in this specification, manages the address space of the MBS, and manages the nodes in consideration of the available space of the MBS available in each node. When an MBS allocation request occurs, storage space is allocated by selecting an MBS that can achieve high file processing performance according to the set policy. Hereinafter, the role of the storage device management unit 1000 will be described in detail with reference to FIG. 5 .

5 is a configuration diagram of a storage device management unit according to an exemplary embodiment.

Referring to FIG. 5 , the storage device management unit 1000 includes an address management unit 100, a space management unit 200, and a monitoring unit 300.

For explanation, the address management unit 100, the space management unit 200, and the monitoring unit 300 are named and called, but they are computing devices operated by at least one processor. Here, the address management unit 100, the space management unit 200, and the monitoring unit 300 may be implemented in one computing device or distributedly implemented in separate computing devices. When distributed and implemented in separate computing devices, the address management unit 100, the space management unit 200, and the monitoring unit 300 may communicate with each other through a communication interface. The computing device suffices as long as it is capable of executing a software program written to perform the present invention, and may be, for example, a server, a laptop computer, or the like.

The address management unit 100 serves to distinguish MBS and DRAM connected to the memory bus, designate and manage the node number where the MBS is installed. A detailed role of the address management unit 100 will be described with reference to FIG. 7 .

The space management unit 200 manages MBS resources through allocation and recovery, determines the size of MBS free space available in each node, and classifies and manages candidate nodes having free space larger than a reference value and obese nodes that do not. there is.

The space management unit 200 allocates an MBS at a location where fast file writing is possible according to an MBS allocation policy. The MBS allocation policy performed by the space management unit 200 may be determined through simulation. Specifically, the relationship between the distance between the CPU and the MBS according to the file processing size (block) is profiled, and based on the collected data, it is determined which node's MBS storage space is allocated and used to achieve high file processing performance. can be modeled.

For example, in a NUMA system with 4 nodes or 8 nodes, 40 processes on only the CPU on one node perform random writes with various block sizes, and the number of nodes on which the processes are running and on which MBS they use is You can see the difference in performance depending on the location.

According to the experimental results, the highest performance can be obtained when the node where the process is running and the node where the MBS used is installed are the same, and the lowest performance can be obtained when using the MBS installed on another node.

On the other hand, even when using MBSs installed on different nodes, it can be seen that higher file processing performance is shown when MBSs installed on multiple nodes are sequentially used than when MBSs on one remote node are used.

Accordingly, the space management unit 200 may reflect this result to the MBS allocation policy to set a local policy for allocating the MBS space of the same node where the process operates and a striping policy for sequentially allocating MBSs installed in multiple nodes. A detailed operation method of the allocation policy will be described with reference to FIGS. 8 to 10 .

The monitoring unit 300 periodically monitors the size of the free space of the candidate node and the obese node classified by the space management unit 200, and changes the state of the node according to the change in the size of the free space. A detailed operating method of the monitoring unit 300 will be described with reference to FIG. 11 .

Meanwhile, the storage device management unit 1000 may further include an interface driver (not shown) for linking with a file system.

6 is an exemplary diagram illustrating a structure in which an MBS is installed in a NUMA system according to an embodiment.

Referring to FIG. 6 , a method of installing MBS in a NUMA system may vary and is not limited to any one.

As shown in (a) of FIG. 6, the MBS may be installed on only one node out of n nodes, or the MBS and DRAM may be installed together on only a few nodes as shown in (b) of FIG. Meanwhile, in this specification, as shown in (c) of FIG. 6, it is assumed that MBS and DRAM are installed in all n nodes. Therefore, in the following specification, 'node' means a node in which both MBS and DRAM are installed.

7 is an explanatory diagram of an operating method of an address management unit according to an exemplary embodiment.

Referring to FIG. 7 , devices used in a computer system may receive and use an address space through which an operating system can access a corresponding device based on installation information. At this time, MBS and DRAM are installed on the same memory bus in terms of hardware and the same physical address is assigned.

Therefore, devices connected to the memory bus cannot be distinguished only with physical addresses, and information on devices installed on the memory bus is additionally required to distinguish between MBS and DRAM.

The address management unit 100 is configured to be used in one logical address space using MBS installation information that is not physically installed. As an example of how to obtain installation information, it can be obtained through the e820_entry structure of e820map during the booting process, and information on the start address, size, and type can be collected. A type value may be used to distinguish between MBS and DRAM. Installation information obtained through the e820_entry structure in the device installed as shown in FIG. 6 may be shown in Table 1.

physical address A B X Y size sizeA sizeB sizeX sizeY type DRAM MBS DRAM MBS

On the other hand, the node number where MBS is installed is information dependent on the system structure, and cannot be obtained through the e820_entry structure, but can be obtained using functions provided by the used system. As an example, the MBS physical address range can be obtained using memory affinity information of the system of ACPI (Advanced Configuration and Power Interface). Through this, the address manager 100 can know which node the MBS belongs to and can designate the node number where the MBS is installed.

The address management unit 100 may manage the address where the MBS is installed for each MBS, the size of the MBS, the node number where the MBS is installed, the total size of the MBS, and a list of MSBs in a structure.

8 is an explanatory diagram of a method for managing a node by a space management unit according to an embodiment.

Referring to FIG. 8 , the space manager 200 divides and manages each node into a candidate node and an obese node according to the size of MBS free space available in each node.

A candidate node means a node having an MBS extra space larger than the reference value, and an obese node means a node having an MBS extra space smaller than or equal to the reference value.

The reference value can be set by a user or administrator, and can be expressed as a ratio of free space to the total size of the MBS installed in each node. As an example, as shown in FIG. 10 , it is assumed that a 10 GB MBS is installed in each node and the reference value is set to 0.1. Based on 1GB, which is 10% of 10GB, nodes with more than 1GB of free space on each node become candidate nodes, and nodes that do not become obese nodes.

As another example, when the reference value is set to 0, an obese node means a node without free space, and nodes with free space may be candidate nodes.

9 is a flowchart of a method for managing a node by a monitoring unit according to an embodiment.

Referring to FIG. 9 , the MBS free space of a node may be changed in real time, and the monitoring unit 300 determines the candidate node and the obese node according to the MBS free space state of the node.

When candidate nodes and obese nodes are classified by the space management unit 200, the monitoring unit 300 periodically monitors the MBS free space size of each node. Determines whether a node is a state.

The monitoring unit 300 compares the MBS free space of a random node i with a reference value (S110).

In step S110, if the free MBS space of node i is greater than the reference value, the monitoring unit 300 checks whether node i is set as a candidate node (S120).

If node i is set as a candidate node, it is set according to the size of the free space, so the step ends (S140).

If node i is not set as a candidate node in step S120, the monitoring unit 300 changes node i from an obese node to a candidate node (S130).

In step S110, if the free MBS space of node i is smaller than the reference value, the monitoring unit 300 checks whether node i is set as an obese node (S150).

If node i is set as an obese node, it is set according to the size of the free space, so the step ends (S140).

If node i is not set as an obese node in step S150, the monitoring unit 300 changes node i from a candidate node to an obese node (S160).

Meanwhile, the monitoring unit 300 may operate as a kernel thread, and may be in a sleep state when there is no obese node. When an obese node occurs, the monitoring unit 300 may wake up and operate.

10 is an explanatory diagram of a method for allocating an MBS by a space management unit according to one embodiment, and FIG. 11 is an explanatory diagram of a method for allocating an MBS by a space management unit according to another embodiment.

As described in FIG. 5, through simulation, the highest file processing performance can be obtained when the process and the MBS used are on the same node, and when the MBS of each node is used sequentially (striping), the MBS installed on the remote node is used. It can be seen that higher file processing performance can be obtained than when using only .

The space management unit 200 may provide two policies to maximize the processing performance of the MBS based on the simulation results.

Referring to FIG. 10 , the space management unit 200 may perform a local policy for allocating space in the MBS of the same node having the smallest access delay time in the node where the process operates.

For example, when a process running on node 0 requests MBS storage space allocation, the space manager 200 may allocate MBS space on the same node where the corresponding process is executed according to a local policy.

On the other hand, the local policy can obtain the highest file writing performance, but there is a problem that it cannot be used after the MBS resource of the corresponding node is exhausted. Therefore, we need a policy that can be applied when the local policy is not available.

Referring to FIG. 11 , the space management unit 200 may perform a striping policy of sequentially allocating MBSs among assignable MBSs.

For example, when a process running on node 0 requests allocation of MBS storage space, the space management unit 200 may first allocate the MBS space installed on node 0 and then allocate the MBS space installed on node 1.

On the other hand, the striping policy has lower file writing performance than the local policy, but higher file writing performance can be obtained than when using only one remote node. In addition, since the MBS of each node can be evenly used, exhaustion of only the MBS resources of a specific node can be prevented.

When MBSs of multiple nodes are striped, higher file processing performance can be obtained than when MBSs of nodes with the lowest access latency excluding local nodes are used. The striping technique can be used as a technique to obtain the highest file processing performance under conditions where the local technique cannot be used.

Which policy the space management unit 200 will use as a basic policy can be set by a user or an administrator, and can be set at compile time of the space management unit 200 . As an example, a user or administrator may select a policy when executing an application through an I/O Control interface. If the policy to be used is not selected, the default policy can be used.

The space management unit 200 may manage whether the entire storage space is used in various ways, and as an example, a bitmap may be used. A bitmap is a method of dividing a space into a certain size and displaying each space with one bit. In order to find available space, the bitmap must be sequentially searched, and in some cases all bits must be searched, and NUMA can be difficult to recognize.

As another example, the space management unit 200 may use a buddy system. A buddy system refers to a system in which memory is allocated in units of powers of 2 (4KB, 8KB, 16KB...) from segments.

The buddy system enables faster search and management compared to bitmap, and has been used for a long time in the main memory usage environment. Since the MBS has the same characteristics as the main memory, the space management unit 200 can allocate or recover the physical space of the MBS through the pseudo code shown in Table 2 using the buddy system.

Function Get Striping(node id, order)
return GetMBSfromallNodes ;
end
Function Get Local(node id, order)
if Local MBS has enough Spacethen
return Get M BS from N ode(node id) with order ;
else
return GetStriping(node id, order) ;
end
end
Function Apply (Policy, order)
if Policy != LOCALthen
return Get Striping(processnid, order) ;
else
return Get Local(processnid, order) ;
end
end
FunctionGetPolicy(currentprocess)
if currentprocesspolicy != NULLthen
return currentprocess_policy ;
else
return NUMA - store_DefaultPolicy ;
end
end
Function alloc_mbs(order)
Policy = GetPolicy(currentprocess) ;
Apply(Policy) ;
end

In Table 2, order means the number of MBS pages to be allocated, processnid means the node number on which the process requesting MBS allocation is running, processpolicy means the MBS allocation policy desired by the process, and NUMA - store_DefaultPolicy means the default allocation policy set at compile time.

12 is a flowchart of a method of managing a node by a storage manager according to an embodiment.

Referring to FIG. 12 , the address management unit 100 of the storage device management unit 1000 collects installation information of devices connected to the memory bus (S210). As an example, it can be obtained through the e820_entry structure of the e820map during the booting process, and information on the start address, size, and type can be collected.

The address management unit 100 of the storage device management unit 1000 distinguishes between MBS and DRAM using the collected installation information (S220). As an example, collected information can be classified as a type value.

The address management unit 100 of the storage device management unit 1000 collects the MBS physical address range from memory affinity information of the system (S230). As an example, memory affinity information of an Advanced Configuration and Power Interface (ACPI) system may be used.

The address management unit 100 of the storage device management unit 1000 designates the node number where the MBS is installed based on the MBS physical address range (S240).

The address management unit 100 of the storage device management unit 1000 manages the node where the MBS is installed by dividing the node into a candidate node and a non-existent node (S250). Thereafter, the monitoring unit 300 periodically monitors the MBS free space size of each node, and when the free space size changes, it can change whether each node is a candidate node or an obese node according to a reference value.

13 is a flowchart of a method of allocating a storage space by a storage device manager according to an exemplary embodiment.

Referring to FIG. 13 , the storage device manager 1000 receives a file write request (S310).

The storage device management unit 1000 checks what the default allocation policy of MBS is set to (S320). The default allocation policy can be a local policy or a striping policy.

If the default allocation policy is the local policy in step S320, the storage device management unit 1000 checks whether there is free space in the local node (S330). A local node means a node where a process operates.

If there is free space in the local node, the storage device management unit 1000 allocates storage space in the local MBS (S340). A local MBS is an MBS that belongs to the node where the process is running.

If there is no free space in the local node, the storage device management unit 1000 checks whether there are other candidate nodes other than the local node (S360). A candidate node may refer to a node having an MBS free space larger than a reference value.

If there is a candidate node, the storage management unit 1000 allocates storage space in the candidate node (S370), and if there is no candidate node, the storage device management unit 1000 allocates space in the obese node (S380). Meanwhile, if there is no free space in the obese node, the step may be terminated due to memory shortage (ENOSPC).

If the default allocation policy is not set in step S320, the storage device management unit 1000 checks whether the node to which the local MBS belongs is an obese node (S350).

If the node to which the local MBS belongs is not an obese node, the storage manager 1000 allocates storage space in the local MBS (S340).

If the node to which the local MBS belongs is an obese node, the storage manager 1000 checks whether there are other candidate nodes besides the local node (S360).

Then, if there is a candidate node, the storage manager 1000 allocates a storage space in the candidate node (S370), and if there is no candidate node, allocates a space in the obese node (S380).

14 is a hardware configuration diagram of a computing device according to an embodiment.

Referring to FIG. 14 , the address management unit 100, the space management unit 200, and the monitoring unit 300 execute instructions described for executing the operation of the present invention in the computing device 400 operated by at least one processor. Execute the program containing the (instructions).

Hardware of the computing device 400 may include at least one processor 410, memory 420, storage 430, and communication interface 440, and may be connected through a bus. In addition, hardware such as an input device and an output device may be included. The computing device 400 may be loaded with various software including an operating system capable of driving programs.

The processor 410 is a device for controlling the operation of the computing device 400, and may be various types of processors that process commands included in a program, for example, a Central Processing Unit (CPU) or a Micro Processor Unit (MPU). ), MCU (Micro Controller Unit), GPU (Graphic Processing Unit), and the like. Memory 420 loads a corresponding program so that the instructions described to carry out the operations of the present invention are processed by processor 410 . The memory 420 may be, for example, read only memory (ROM) or random access memory (RAM). The storage 430 stores various data, programs, etc. required to execute the operation of the present invention. The communication interface 440 may be a wired/wireless communication module.

According to the present invention, since a file processing path is simplified in consideration of MBS characteristics in the Linux kernel, performance loss due to unnecessary overhead can be prevented and file processing performance can be improved.

In addition, according to the present invention, since storage space can be allocated from the MBS of a node capable of minimizing an access delay time, system performance can be improved.

The embodiments of the present invention described above are not implemented only through devices and methods, and may be implemented through programs that realize functions corresponding to the configuration of the embodiments of the present invention or a recording medium on which the programs are recorded.

Although the embodiments of the present invention have been described in detail above, the scope of the present invention is not limited thereto, and various modifications and improvements made by those skilled in the art using the basic concept of the present invention defined in the following claims are also included in the scope of the present invention. that fall within the scope of the right.

Claims (13)

A method of operating a computing device operated by at least one processor,
Collecting installation information including physical addresses of storage devices connected to a memory bus, determining nodes in which the storage devices are installed based on the installation information, and managing free storage space of each node in which the storage devices are installed;
Receiving a file write request and a storage space allocation request according to the file write request from a process executed on an arbitrary node; and
Allocating space in a storage device belonging to the arbitrary node in which the process is executed or allocating space in a storage device belonging to a node different from the arbitrary node according to a preset storage space allocation policy.
Including, operating method. A method of operating a computing device operated by at least one processor,
Managing the free storage space of each node in which the storage device is installed;
Receiving a file write request and a storage space allocation request according to the file write request from a process executed on an arbitrary node; and
Allocating space in a storage device belonging to the arbitrary node where the process is executed or allocating space in a storage device belonging to a node different from the arbitrary node according to a preset storage space allocation policy,
The management step is
Dividing each of the nodes into a candidate node having a storage margin larger than a preset reference value and an obese node having a storage margin equal to or smaller than the reference value. delete In paragraph 1,
The collecting step is
Collecting the physical address where the storage device is installed using the physical address and memory affinity information of the system. In paragraph 1,
The computing device operates as a system with a Non-Uniform Memory Access (NUMA) structure,
The storage device is a memory storage device (Memory Bus connected Storage, MBS) used by connecting to a memory bus, operating method. In paragraph 1,
The storage space allocation policy,
Any one of a local policy for allocating space on a storage device belonging to the arbitrary node and a striping policy for sequentially allocating space on a storage device belonging to a plurality of nodes including the arbitrary node. In paragraph 6,
The storage space allocation policy,
In the process of executing an application through an input/output control (I/O Control) interface of the computing device, the operating method is set to either the local policy or the striping policy. A method of operating a computing device operated by at least one processor,
Managing storage space of each node including a storage device and a central processing unit;
Receiving a storage space allocation request from a process executed on an arbitrary node; and
Allocating space in a storage device belonging to a node with the shortest access time from the arbitrary node where the process is executed,
The computing device is a non-uniform memory access (Non-Uniform Memory Access, NUMA) structured system, operating method. In paragraph 8,
The management step is
Monitoring the free storage space of each node, setting a node having a free storage space larger than a preset reference value as a candidate node, and setting a node having a free storage space smaller than or equal to the reference value as an obese node. how it works. In paragraph 9,
The storage device is a memory storage device (Memory Bus connected Storage, MBS) used by connecting to a memory bus,
The central processing unit includes a plurality of cores (Core), operating method. In paragraph 8,
The allocating step is
Allocating space in a storage device belonging to the arbitrary node. In paragraph 8,
The allocating step is
Checking free storage space of a storage device belonging to the arbitrary node; and
If there is no free storage space in the storage device belonging to the arbitrary node, sequentially allocating the storage space of the arbitrary node and nodes having a short access time among other nodes.
Including, operating method. In paragraph 8,
The allocating step is
Using a buddy memory allocation method of repeatedly dividing the free storage space by 2 and allocating a memory having a size closest to the size of the requested storage space.

Images