KR102044775B1 - Method for User based Resource Scheduling under Multi-User Environment - Google Patents

Abstract

In a multi-user environment, a user-based resource scheduling method includes allocating an individual resource group for each user based on a user identifier (UID), and recalling all processes corresponding to the application when the user executes the application. Automatically assigning to individual user resource groups. User-based resource scheduling can be created by creating user-specific resource groups instead of resource groups previously created by processes. Resource groups are isolated from groups of other users, enabling performance isolation between users.

Description

Method for User based Resource Scheduling under Multi-User Environment}

The present invention relates to a computing system resource scheduling technique, and more particularly, to a surface computing system resource scheduling technique in a multi-user environment.

Surface computing with a multi-user environment requires resource reservations through strict criteria for resource requests to ensure that each user's reliable performance provides timely reserved resources. In other words, user-based scheduling and performance isolation between users are required.

For strict resource reservation, the operating system performs scheduling analysis on the resources requested by real-time tasks. Linux provides this resource reservation using the control group framework (cgroup). The cgroup provides resources such as a CPU, a memory, a network, and the like through a subsystem corresponding to each resource, and each resource is provided in a task group unit.

However, the latest Linux system related to the prior art allocates CPU resources based on ttys because cgroups are created for each tty (Teletypewriter), which means a console or a terminal among Linux device drives. Here cgroups have the ability to limit and isolate the resource usage of processes in the Linux kernel. In this case, CPU resource scheduling and / or I / O resource scheduling in a modern Linux system can provide performance isolation and fairness between ttys, but cannot provide fair resource allocation and performance isolation between users.

Also, in order for existing cgroups to be used for user-based scheduling, there is a problem that each process needs an additional process migration from the existing tty-specific group to the user-specific group, and it must be able to track ownership information of all processes. In modern modern Linux systems, if you create per-user groups via the cgroup interface for inter-user fairness, performance isolation, and weight-based assignments, the autogroup-to-user group is automatically created when the process is created. Unnecessary migration of task groups occurs. That is, the task group should be removed from the existing tty autogroup and moved to the corresponding user group. In the background application, the user identifier (userid (uid)) and process identifier (process id (pid)) Since there is a need for monitoring, there is a problem in that an overhead occurs.

In addition, conventional user-based scheduling-related techniques, such as the Fair Share Scheduler (FSS) scheduler, focus only on fairness between users, thereby limiting the inability to allocate more resources to desired users.

Patent Publication No. 10-2009-0055018 (Permission Management System, International Business Machines Corporation) Korean Patent Publication No. 10-1733534 (Performance Analysis Method and Apparatus for Hard Real-Time Scheduling Task Group, Agency for Defense Development)

Task Group-based Scheduling Scheme for Real-Time Considerations in TinyOS in Non-Preemptive Environments, Journal of Korea Multimedia Society, 2010.09, vol 13, pages 1285-1298

An object of the present invention for solving the above problems is to allocate the central processing unit (CPU) resources and / or input / output (I / O) resources based on the user to provide isolation of the central processing unit and / or input / output performance between users To provide a user-based resource scheduling method in a multi-user environment.

According to an aspect of the present invention for solving the above problems, the user-based resource scheduling method in a multi-user environment, the step of assigning a separate resource group for each user based on a user identifier (UID), Automatically allocating all processes corresponding to the application to the user's respective resource group when the user executes the application. The resource scheduling method may be a central processing unit (CPU) resource scheduling method. The CPU resource scheduling method may include allocating a user task group for each user identifier and automatically allocating processes having the same user identifier to the user task group. The CPU resource scheduling method may further include assigning a weight based on the user identifier and differentially providing the CPU resource to the user based on the weight. In the CPU resource scheduling method, as the weight increases, the virtual execution time increases more slowly than the actual execution time, and as the weight decreases, the virtual execution time increases faster than the actual execution time. . The CPU resource scheduling method allocates more time by allowing a higher weighted process or a higher weighted user task group to increase the virtual execution time more slowly than a lower weighted process or a lower weighted user task group. Resources can be scheduled to receive. The CPU resource scheduling method may further include scheduling to be equally distributed to the process being used based on the user identifier. The CPU resource scheduling method may further include limiting the use of the CPU resource to a malicious user to a predetermined value or less. In the CPU resource scheduling method, a process and a user task group having the lowest virtual execution time may be selected and scheduled. The CPU resource scheduling method may select a tty task group having the lowest virtual execution time from the selected user task group, and select a process having the lowest virtual execution time from the selected tty task group. The CPU resource scheduling method may re-request the CPU resource scheduling when all of the maximum virtual execution time that the selected process can execute or when the selected process is stopped. The resource scheduling method may be an input / output (I / O) resource scheduling method. The I / O resource scheduling method may include allocating a user I / O group for each user identifier and allocating processes having the same user identifier to the same user I / O group. The input / output (I / O) resource scheduling method may further include assigning a weight based on the user identifier and differentially providing the input / output (I / O) resource to the user based on the weight. In the I / O resource scheduling method, a user I / O group having the lowest virtual disk time is selected and scheduled. The higher the weight, the slower the virtual disk time increases than the actual time used, and the user I / O having a high weight. Resources may be scheduled so that groups are allocated more time by allowing the virtual disk time to increase more slowly than user I / O groups with lower weights. The input / output (I / O) resource scheduling method may further include scheduling to distribute the input / output (I / O) resources evenly to a process being used based on the user identifier. In the I / O resource scheduling method, a user I / O group having the lowest virtual disk time may be selected and scheduled. The input / output (I / O) resource scheduling method may select a process having the highest priority from the selected user input / output group. The input / output (I / O) resource scheduling method includes the real time (RT) process, the best effort process, and the idle (IDLE) process, the real time (RT) process, the The priority may be higher in order of best effort process and IDLE process. The I / O resource scheduling method may schedule the user I / O group having the lowest virtual disk time to be selected between the user I / O groups, and the process or the process I / O queue in the user I / O group may be scheduled in a round robin manner. .

According to a user-based resource scheduling method for supporting a multi-user environment according to embodiments of the present invention, a user task group, which is an individual resource group, is assigned to each user based on a user identifier (userid (uid)), and the user processes By automatically assigning a process to a corresponding user task group when creating, it is possible to support user-based resource scheduling by creating a user-specific resource group instead of a tty (Teletypewriter) or a resource group previously created by a process. It is isolated from groups of users to support performance isolation between users. In this case, unlike the cgroup interface of the existing Linux system that can provide a function similar to the present invention, since the user group is created before the tty and process resource groups are created, the process migration as described above is performed. No overhead is incurred.

In addition, according to the user-based resource scheduling method for supporting a multi-user environment according to an embodiment of the present invention, a task group and / or input and output resource group (CFQ group) which is a central processing unit resource group used for cgroup Modifications created per user allow for fair resource allocation per user and isolation of CPU and / or I / O performance.

In addition, according to the user-based resource scheduling method for supporting a multi-user environment according to an embodiment of the present invention, rather than the same amount of resource allocation, the weight of each resource group is adjusted to support resource allocation according to a desired ratio of the system By adjusting the weights of task group and I / O group objects, weight-based resource allocation can be supported.

1 is a flowchart illustrating a user-based resource scheduling method for supporting a multi-user environment according to an exemplary embodiment of the present invention.
2 is a flowchart illustrating a user-based central processing unit resource scheduling method for a surface computing system supporting a multi-user environment according to an embodiment of the present invention.
3 is a conceptual diagram illustrating a user-based central processing unit resource scheduling method for a surface computing system supporting a multi-user environment according to an embodiment of the present invention.
4 is a flowchart illustrating a user-based input / output (I / O) resource scheduling method for a surface computing system supporting a multi-user environment according to another embodiment of the present invention.
5 is a conceptual diagram illustrating a user-based input / output (I / O) resource scheduling method for a surface computing system supporting a multi-user environment according to another embodiment of the present invention.
6 illustrates a CPU processing method according to the prior art for comparing with a user-based resource scheduling method in a multi-user environment according to embodiments of the present invention.
7 illustrates a user-based central processing unit (CPU) scheduling scheme according to another embodiment of the present invention.
FIG. 8 illustrates a CPU scheduling method using user-based weights according to another embodiment of the present invention.
9A illustrates a user-based CPU resource fairness experiment result according to an embodiment of the present invention.
9B illustrates a result of a user-based central processing unit resource performance isolation experiment according to an embodiment of the present invention.
9C illustrates a result of a user-based input / output (I / O) resource process experiment according to an embodiment of the present invention.
9D illustrates a result of a user-based resource performance isolation experiment according to an embodiment of the present invention.

1 is a flowchart of a user-based resource scheduling method for a surface computing system supporting a multi-user environment according to an exemplary embodiment of the present invention.

Referring to FIG. 1, in a user-based resource scheduling method supporting a multi-user environment according to an exemplary embodiment of the present invention, allocating individual resource groups for each user based on a user identifier (UID) (S110). And automatically assigning all processes corresponding to the application to the individual resource group of the user (S120) when the user executes the application.

Here, the user identifier may identify a user by allocating a user task group in the CPU processing method of the surface computing system supporting the multi-user environment. Alternatively, the user identifier may identify a user by allocating a user based input / output group (User CFQ Group) in an input / output (I / O) resource scheduling method in a surface computing system supporting a multi-user environment.

A user-based central processing unit resource scheduling method for a surface computing system supporting a multi-user environment will be described in detail with reference to FIGS. 2 and 3, and user-based input / output (I / O) resources for a surface computing system supporting a multi-user environment are described below. The scheduling method will be described later with reference to FIGS. 4 and 5.

A task group is used as a CPU resource group in the real kernel, and the kernel manages this object in a parent-child hierarchy. In the old Linux kernel, the setsid system call is called whenever a new terminal is executed and a tty is created, and a task group for that tty is created to create a new child of the root task group. Become a group. Here, tty is usually a terminal (session) in which a user communicates with a computer in a Linux system, and setsid is a Linux system call that creates a session and specifies a program group ID.

However, in the user-based resource scheduling method in a multi-user environment according to the embodiments of the present invention, instead of generating a task group for each tty, a user task group for each user identifier (uid) is generated and the same user identifier (uid) is generated. Assign or insert processes with) into the same user task group. Since the user task group is independent of other user task groups, the user-based resource scheduling method in the multi-user environment according to the embodiments of the present invention enables performance isolation by user. In addition, in a user-based resource scheduling method in a multi-user environment according to embodiments of the present invention, a unique weight may be adjusted for each user task group, and thus resources may be allocated to the user based on the weight. Hereinafter, the resource means a central processing unit resource or an input / output (I / O) resource.

For example, if User 1 is more important when there are Users 1, 2, and 3, a weight can be given to User 1 in terms of the system. For example, if the weight is given to each user 2: 1: 1, users 1, 2, and 3 will be allocated resources of 50%, 25%, and 25%, respectively.

In a multi-user environment, the user-based resource scheduling method according to the embodiments of the present invention allocates or inserts a user task group immediately when a process is created, and thus, overhead due to migration between additional task groups. Does not occur.

2 is a flowchart illustrating a user-based central processing unit resource scheduling method for a surface computing system supporting a multi-user environment according to an embodiment of the present invention.

Referring to FIG. 2, in the user-based CPU resource scheduling method for a surface computing system supporting a multi-user environment according to an embodiment of the present invention, first, a user-based CPU resource scheduling request is determined (S200). do.

Since each user has a separate user task group, user-specific scheduling is possible.

When there is a request for user-based CPU resource scheduling, the user task with the lowest virtual runtime (vRuntime) among the core task groups is searched by searching the queue with the minimum virtual runtime (vRuntime). Select a group (User Task Group) (S210). Here, vRuntime is an abbreviation of virtual runtime, and is a value in which a process uses a CPU in consideration of the overall process and priority of the process.

In the selected user task group, the tty (Teletypewriter) task group having the lowest virtual execution time is selected (S220), and the process having the lowest virtual execution time is selected from the selected tty task group (S230).

Increase the virtual execution time of the process and the group to which the process is assigned, where the group means a user task group or a tty task group (S240), and all of the maximum virtual execution times that the process can run. If the corresponding process is stopped or used, it requests for central resource scheduling again (S250). Here, the speed of increase of the virtual execution time may be changed by a weight that is adjusted or set by an administrator to be described later. For example, higher weights increase the virtual execution time more slowly than the actual execution time. That is, the higher the weight, the virtual execution time increases more slowly than the actual execution time, and the lower the weight, the virtual execution time increases faster than the actual execution time. Therefore, it is possible to implement such that resources are scheduled such that a higher weight process (or user task group) is allocated more time by causing the virtual execution time to increase more slowly than a lower weight process (or user task group).

Details of each step of FIG. 2 will be described later with reference to FIG. 3.

3 is a conceptual diagram illustrating a user-based central processing unit resource scheduling method for a surface computing system supporting a multi-user environment according to an embodiment of the present invention.

Referring to FIG. 3, user-based CPU resource scheduling for a surface computing system supporting a multi-user environment according to an embodiment of the present invention achieves fairness based on a virtual runtime of a user task group. Scheduling method.

As described above with reference to FIG. 2, in the CPU resource scheduling method according to an embodiment of the present disclosure, a process and / or a user task group having the lowest virtual execution time may be selected and scheduled.

Referring to FIG. 3, a user-based central processing unit resource scheduling process for a surface computing system supporting a multi-user environment according to an embodiment of the present invention may include a user task group 315-1 of a first user. ),… , User Task Group 315-k of the k-th user,... Assuming that the virtual runtime vRuntime of the user task group 315-k of the k-th user is the lowest among the task groups 315-n of the n-th user, The k-th user's task group 315-k having the lowest virtual runtime vRuntime among the task groups 305 is selected (see S310 and S210).

First teletyper task group 325-1,... J j tty Task Group 325-j,. The virtual runtime (vRuntime) of the j-th teletyper task group (325-j) is the lowest among the n-th typewriter task group (325-n) (ie Assuming that the increase rate of the virtual execution time of the j-th teletypewriter task group 325-j increases most slowly among the n teletypewriter task groups compared to the actual execution time), the k-th user task group 315 the jth teletypewriter task group at the lowest vRuntime (i.e., the rate of increase of the virtual execution time increases most slowly among the n teletypewriter task groups relative to the actual execution time) at -k). (tty Task Group) 325-j is selected (see S320 and S220).

First process 335-1,... , M process 335-m,... Among the nth process 335-n, the mth process 335-m has the lowest virtual execution time vRumtime (that is, the increase rate of the virtual execution time of the mth process 335-m is actually lower than that of the nth process 335-n. The slowest increase among the n processes compared to the execution time), the mth process having the lowest vRuntime in the selected jth tty Task Group 325-j. 335-m is selected (see S330 and S230).

Thereafter, the virtual execution time of the m process 335-m which has the lowest value of the selected virtual runtime vRuntime (that is, the median speed of the virtual execution time increases slowly compared to the actual execution time) is increased, and When the process m (335-m) runs out or stops to the maximum, the user-based central processing unit resource scheduling is re-requested (S340) and is performed again from step S210.

Alternatively, user-based CPU resource scheduling for a surface computing system supporting a multi-user environment according to another embodiment of the present invention may achieve fairness based on a weight of a user task group.

The user task group is weighted by an administrator. In the CPU resource scheduling method according to another embodiment of the present invention, the higher the weight, the slower the virtual execution time is compared to the actual execution time, and the higher the weight (and / or the user task group) the lower the weight. Since the virtual runtime grows more slowly than the process (and / or user task group), resources can be scheduled to allocate more time.

According to another embodiment of the present invention, a user-based central processing unit resource scheduling process for a surface computing system supporting a multi-user environment includes a user task group 315-1,. , User Task Group 315-k of the k-th user,... The weight of the k-th user's task group 315-k is highest among the n-th user's task group 315-n (ie, the k-th user's task group). Assuming that the medium acceleration rate of virtual execution time increases most slowly among the task groups of n users compared to the actual execution time, the core weight of the Core Task Group 305 is the highest (ie, the virtual execution time). The k-th user's task group 315-k may be selected.

First teletyper task group 325-1,... J j tty Task Group 325-j,. The weight of the j th teletype writer task group 325-j is the largest among the n th typewriter task group 325-n (ie, the j th typewriter). The weighting rate in the k-th user task group 315-k is assumed to be the middle acceleration rate of the virtual execution time of the task group 325-j increases most slowly among the n teletypewriter task groups compared to the actual execution time. Is the largest (that is, the medium acceleration speed of the virtual execution time is the slowest increase among the n teletypewriter task groups compared to the actual execution time), the j-th teletypewriter task group 325-j is Can be chosen

First process 335-1,... , M process 335-m,... Among the nth processes 335-n, the m th process 335-m has the largest weight (that is, the middle acceleration rate of the virtual execution time of the m th process 335-m is n compared to the actual execution time). The slowest among the processes), the m th process 335-m having the largest weight may be selected from the selected j th tty Task Group 325-j (S330). , S230).

Then, the virtual execution time of the m process 335-m with the largest weight selected (i.e., the middle speed of the virtual execution time increases most slowly compared to the actual execution time) is increased, and the m process 335-m is increased. When all of the maximum time that can be executed or stopped is used, user-based CPU resource scheduling may be re-requested (S340) and the process may be performed again from step S210.

4 is a flowchart illustrating a user-based input / output (I / O) resource scheduling method for a surface computing system supporting a multi-user environment according to another embodiment of the present invention.

Referring to FIG. 4, in the user-based CPU scheduling method for a surface computing system supporting a multi-user environment according to another embodiment of the present invention, it is determined whether a user-based user input / output (I / O) scheduling request is performed ( S400). Here, since each user has a separate user input / output group (User CFQ Group), scheduling is possible for each user. As a result of the determination, when the user input / output (I / O) scheduling request is received, a virtual disk time (vdisktime) of the device CFQ group is searched by searching a queue having a minimum virtual disk time (vdisktime). The lowest user input / output group (User CFQ Group) is selected (S410). A service tree or a process having the highest priority is selected from the selected user input / output group (S420). Here, the priority may be higher in order of a real time (RT) process 521, a best effort (BE) process 522, and an idle (IDLE) process 523.

In the selected service tree or process, an CFQ queue is selected for each CFQ queue in a round robin manner (S430). In this case, according to the round robin method, each input / output queue (CFQ Queue) for a certain time-during a specific time slice-to send I / O data and when the corresponding time slice ends the corresponding input / output sequentially It is scheduled by selecting the input / output queue (CFQ Queue) next to the queue (CFQ Queue). In other words, the user CFQ group is scheduled to select the user CFQ group having the lowest virtual disk time (vdisktime), and the process (or process I / O queue) within the user CFQ group is selected. Process CFQ queue)) is scheduled in a round robin fashion.

After increasing the virtual disk time of the selected user CFQ group (S440), if all the virtual disk time is consumed, the I / O resource is allocated to another user CFQ group having the lowest virtual disk time. do. Details for each step will be described later with reference to FIG. 5.

As described above with reference to FIG. 4, in the user-based input / output (I / O) resource scheduling method according to an embodiment of the present invention, the user CFQ group having the lowest virtual disk time is selected. The higher the weight, the slower the virtual disk time is compared to the actual time used, and the higher the weight of the user I / O group (User CFQ Group) than the lower weight I / O group (User CFQ Group). As it grows more slowly, resources can be scheduled to allocate more time.

5 is a conceptual diagram illustrating a user-based input / output (I / O) resource scheduling method for a surface computing system supporting a multi-user environment according to another embodiment of the present invention.

Referring to FIG. 5, a user CFQ group for each user identifier is allocated, and a process CFQ queue corresponding to all processes created by the user is allocated to the user I / O group. Here, the I / O queue is an I / O queue managed by the Linux Complete Fair Queuing (CFQ) I / O scheduler, and is an I / O queue allocated to each process individually. Each process has its own input / output queue (CFQ queue) to ensure fairness between processes. I / O group (CFQ group) is the upper layer of the I / O queue (CFQ queue) and is used when scheduling by grouping several processes into one group.

The user I / O group is separately managed through a specific service tree according to the weight or the type of the I / O queue. Here, the type of input / output queue (CFQ queue) may be a real time (RT) process 521, a best effort (BE) process 522, and an idle (IDLE) process 523. Alternatively, the type of the input / output queue may be sync, async, sync-idle, or sync-noidle. A detailed description will be described later with reference to FIG. 5.

5, a first user input / output group (User CFQ Group) 515-1,... , k th User CFQ Group 515-k,... Assuming that the virtual disk time of the k-th user input / output group (515-k) is the lowest among the nth user input / output group (515-n), the virtual disk time (vdisktime) is The lowest k-th user CFQ Group 515-k is selected (S510, see step S410).

The service tree or process having the highest priority in the selected user input / output group is selected (see S420). The process may be at least one of a real time (RT) process 521, a best effort (BE) process 522, and an idle (IDLE) process 523. For example, the priority may be higher in order of real time (RT) process 521, best effort (BE) process 522, and idle (IDLE) process 523.

The real time (RT) process 521 among the real time (RT) process 521, the best effort (BE) process 522, and the idle (IDLE) process 523 is a priority. Is assumed to be the highest, a real time (RT) process 521, which is the highest priority process, is selected in the K-th user I / O group 515-k (see S520 and S420). Here, the real time (RT) process 521 has a higher priority than the BE process 522 and the idle process 523, so the RT process 521 is always the best type. It may be executed before the BE process 522 or the idle process 523.

The type (or type) of I / O sends the I / O data (confirmation) and the process waits for the I / O operation to complete, and the next code without waiting for the I / O operation to complete. There is async to run. If the requested data is synchronous and is sequential, the sync-idle is not sequential. If it is not sequential, it is sync-noidle.If the I / O is async, I / O is queued as async. Is assigned or inserted.

Subsequently, as described above in S430 of FIG. 4, since the process (or process input / output queue) in the user input / output group (User CFQ Group) is scheduled in a round robin manner, the process input / output in a round robin manner in FIG. 5. The process CFQ queue is selected (S530), and the virtual disk time of the k th user I / O group 515-k is determined. To increase (see step S540, step S440). When the virtual disk time is exhausted, I / O resources are allocated to another user CFQ group with the lowest virtual disk time. When CFQ Queue is selected by round robin, if all the time slices are consumed with a specific time slice according to the priority of each CFQ Queue It is scheduled to be next to the input / output queue (CFQ Queue).

6 illustrates a CPU processing method according to the prior art for comparing with a user-based resource scheduling method in a multi-user environment according to embodiments of the present invention.

Referring to FIG. 6, the existing Linux system uses the FSS scheduler to provide the central processing unit and I / O resource scheduling for the purpose of process-specific fairness rather than providing weight-based flexible resource allocation to users. Referring to FIG. 6, the first user 610 is allocated 10% of the CPU resources, the second user 620 is allocated 20% of the CPU resources, and the third user 630 receives 30% of the CPU resources. The fourth user 640 received 40% of the resource. The first to tenth processes all show 10% allocation of the same CPU resource.

If a user-specific group is created through the existing cgroup interface for fairness, performance isolation, and weight assignment between users, unnecessary migration occurs from the autogroup that is automatically entered when the process is created. do.

However, the present invention solves the problems by allocating an individual resource group for the user based on a user identifier (uid) and automatically assigning all processes of the user to the user group when the user executes an application.

7 illustrates a user-based central processing unit (CPU) scheduling scheme according to another embodiment of the present invention.

Referring to FIG. 7, the present invention allocates a user task group based on a user identifier, and assigns processes having the same user identifier to the same user task group. Since the user task group is executed independently from other user task groups, CPU resource scheduling is possible for each user identifier.

Referring to FIG. 7, the first user task group 710, the second user task group 720, and the third user task group 730 and the fourth user task group 740 are surface computing supporting a multi-user environment. In the user environment for the system, CPU resources are allocated 25% each.

7, a first process 711 is assigned to the first user task group 710 and the first process 711 is a centrally assigned to the first user task group 710 alone. The following shows an example in which up to 25% of processor resources are allocated.

Similarly, referring to FIG. 7, the second process 721 and the third process 722 are assigned to the second user task group 720 and have the same user identifier. In addition, the second processor 721 and the third process 722 are allocated a maximum of 12.5% resources of the central processing unit 25% allocated to the second user task group 720, respectively.

In addition, referring to FIG. 7, the fourth process 731, the fifth process 732, and the sixth process 733 are assigned to the third user task group 730 and have the same user identifier. The fourth process 731, the fifth process 732, and the sixth process 733 each allocate up to 8.3% of resources of the central processing unit 25% allocated to the third user task group 730. Show what you received.

7, the seventh process 741, the eighth process 742, the ninth process 743, and the tenth process 744 are assigned to the fourth user task group 740 and have the same user identifier. Has Referring to FIG. 7, resources of the central processing unit 25% allocated to the fourth user task group 740 are allocated to the seventh process 741, the eighth process 742, and the ninth process 743. The tenth process 744 shows an example of being allocated at most 6.25%.

Referring to FIG. 7, the prior art shows a result of supporting per-user fairness and performance isolation by creating a resource group for each user instead of the resource group generated for each tty or process.

According to the present invention, an unnecessary migration problem of the process can be solved by generating a user task group based on the user identifier and automatically allocating processes for each user task group when an application is executed in the user task group. have.

FIG. 8 illustrates a CPU scheduling method using user-based weights according to another embodiment of the present invention.

Referring to FIG. 8, a CPU scheduling method using a user-based weight according to another embodiment of the present invention generates a user task group based on a user identifier, and generates a user task group for each user task group. It provides weight based CPU resource scheduling.

FIG. 8 illustrates user-based scheduling central processor resource weights for a surface computing system supporting a multi-user environment in a first user task group 810, a second user task group 820, and a third user task group 830. An example of assigning weights at 50%, 25%, and 25% ratios, that is, at 2: 1: 1 ratios, respectively.

The first process 811 and the second process 812 are assigned to the first user task group 810 and have the same user identifier. In addition, the first process 811 and the second process 812 may each allocate up to 25% of resources of the central processing unit 50% allocated to the first user task group 810 to use the resources. .

The third process 821 and the fourth process 822 are assigned to the second user task group 820 and have the same user identifier. In addition, the third process 821 and the fourth process 822 may each allocate up to 12.5% of the resources of the central processing unit 25% allocated to the second user task group 820 to use the resources. .

The fifth process 831 and the sixth process 832 are assigned to the third user task group 830 and have the same user identifier. Also, the fifth process 831 and the sixth process 832 are allocated up to 12.5% of resources of the central processing unit 25% allocated to the third user task group 830, respectively.

That is, as shown in FIG. 8, the first process 811 and the second process 812 belonging to the first user task group 810 are allocated to the second user task group 820. Maximum compared with the fifth process 831 and the sixth process 832 assigned to the third process 821, the fourth process 822, and the third user task group 830. Allocate more than twice the CPU resources to use the resources.

9A to 9D illustrate the results of experiments by establishing an environment in which four to five users request resources of the same system to verify that the user fairness and performance isolation between users are provided for the present invention.

In all experiments, Ubuntu 16.04 LTS 64bit operating system was used, and the latest Linux kernel version 4.11 was modified to implement user-specific CPU and I / O resource allocation.

Also, we used mencoder, a video encoding program, and fio, an I / O performance test program, as an I / O benchmark program. In the performance isolation experiment, the stress-ng tool was used to reduce the overall CPU and I / O resource performance by malicious users. All processes were run in an independent terminal.

Referring to FIG. 9A, FIG. 9A illustrates four, one, two, three, four, four CPU processors as a result of a user-based CPU resource fairness experiment for a surface computing system supporting a multi-user environment. The result of mark program execution is shown.

First, using the conventional techniques related to the CPU processing experiment, the second user, the third user, and the fourth user are allocated CPU resources of 2.02 times, 3.04 times, and 4.06 times of the first user. The conventional technique simply allocates resources according to the number of running ttys (teletypewriters) or processes without considering users, so that users are allocated unfair resources proportional to the number of running programs. However, as a result of a user-based CPU resource fairness experiment for a surface computing system supporting a multi-user environment according to embodiments of the present invention, the first user, the second user, the third user, and the fourth user are within 18% of the difference. Will be allocated the same CPU resource. In other words, the conventional technique is process-based, thus providing fairness between processes, so that users with more processes have more total throughput than other users. However, the user-based CPU resource scheduling method for the surface computing system supporting the multi-user environment according to the embodiments of the present invention provides fairness among users so that the total throughput of all processes driven by one user (see FIG. 9A). In this case, the sum of encoding throughputs) is similar between users (within 18%).

Referring to FIG. 9B, FIG. 9B is As a user-based central processor resource performance isolation experiment for a surface computing system supporting a multi-user environment, it shows how a malicious user intentionally affects the performance of other users when performing a large number of processes. The results of five users running the ten CPU stress programs are shown. FIG. 9B illustrates a case where the experimental environment and the number of processes are the same as those of FIG. 9A, but additionally, a malicious user requests a CPU resource by performing a large number of processes. In the CPU resource performance isolation experiment, since the conventional technique does not provide the performance isolation for each user, the overall performance of the system is degraded by a malicious program and adversely affects the performance of existing users. However, according to embodiments of the present invention, since the user owns a unique resource group, the user-specific performance is guaranteed. As a result, according to a user-based central processing unit resource performance isolation experiment for a surface computing system supporting a multi-user environment according to the present invention, the total encoding throughput of the first user, the second user, the third user, and the fourth user is 56% improvement over the prior art. That is, in FIG. 9B, in the user-based CPU processing resource scheduling method according to the embodiments of the present invention, the sum of the total throughputs (encoding throughput in the case of FIG. 9B) of the entire processes of the entire users (first to fourth users) is conventional. Compared to the sum of the total throughputs (encoding throughput in FIG. 9B) of the entire processes of the entire users (first to fourth users) of, are about 56% higher than the conventional scheme. This is because the user-based central processing unit resource scheduling method according to the embodiments of the present invention provides performance isolation between users, thereby reducing performance degradation caused by malicious users compared to the conventional method.

Referring to FIG. 9C, FIG. 8C is a result of a user-based input / output (I / O) resource processing experiment for a surface computing system supporting a multi-user environment, in which four users have one, two, three, four, Shows the results of running the I / O benchmark program. In the conventional technique related to the user-based input / output (I / O) resource processing experiment, the second user, third user, and fourth user allocates 2.11 times, 3.11 times, and 4.15 times I / O resources of the first user. received. According to embodiments of the present invention, users may be allocated a fair resource to allocate resources primarily according to the user. That is, the first user, the second user, the third user, and the fourth user are allocated the same I / O resource within 15% or less.

Referring to FIG. 9D, FIG. 8D is a result of a user-based input / output (I / O) resource performance isolation experiment for a surface computing system supporting a multi-user environment, and the performance of another user when a malicious user intentionally performs a large amount of processes. Figure 9d shows the results of five users running the ten I / O stress programs. Since the present invention owns its own resource group for each of the five users, it ensures performance isolation for each of the five users. That is, according to the present invention, the sum of the read performance throughputs of the first user, the second user, the third user, and the fourth user shows a 66% improvement over the conventional technique. That is, in FIG. 9D, the method for scheduling user-based input / output (I / O) resources according to the embodiments of the present invention is based on the total throughput (read performance throughput in the case of FIG. 9D) of all processes of all users (first to fourth users). Comparing the sum with the sum of the total throughput (read performance throughput in the case of FIG. 9D) of all processes of the conventional users (first to fourth users) is about 66% higher than the conventional method. This is because the user-based input / output (I / O) resource scheduling method according to the embodiments of the present invention provides performance isolation between users, thereby reducing performance degradation caused by malicious users compared to the conventional method.

305: Core task group
315-1: First user task group
315-k: kth user task group
315-n: nth user task group
325-1: First Teletyper Task Group
325-j: jth teletypewriter task group
325-n: nth teletypewriter task group
335-1: First process
335-m: m process
335-1: nth process
505: Device CFQ Group
515-1: User CFQ Group 1st
515-k: kth user input / output group (User CFQ Group)
515-n: nth user input / output group (User CFQ Group)
521: Real-time (RT) process
522: Best Practices (BE) process
523: IDLE process
535-1: Process CFQ Queue
535-j: Process CFQ Queue
535-k: Process m / c queue (Process CFQ Queue)

Claims (20)

In the user-based resource scheduling method performed by the computing system in a multi-user environment, the computing system is
Allocating an individual resource group for each user based on a user identifier (UID); And
Automatically allocating all processes corresponding to the application to an individual resource group of the user when the user executes the application,
The resource scheduling method includes an input / output (I / O) resource scheduling method,
Allocating an individual resource group for each user includes allocating a user input / output group for each user identifier.
The I / O resource scheduling method is
Allocating the user input / output group;
Assigning processes with the same user identifier to the same user input / output group; And
And assigning weights based on the user identifiers to differentially provide the input / output (I / O) resources to the user based on the weights.
delete delete 2. The method of claim 1, wherein the resource scheduling method further comprises a CPU resource scheduling method, wherein the CPU resource scheduling method assigns a weight based on the user identifier and based on the weight. And differentially providing a central processing unit resource to the user. 5. The CPU resource scheduling method of claim 4, wherein the higher the weight, the virtual execution time increases more slowly than the actual execution time, and the lower the weight, the virtual execution time is greater than the actual execution time. A user-based resource scheduling method supporting a multi-user environment characterized by rapid growth. The method of claim 4, wherein the CPU resource scheduling method comprises: a process having a first weight or a user task group having a second weight having a third weight lower than the first weight or the first weight. A method for scheduling user-based resource support in a multi-user environment, wherein resources are scheduled to allocate more time by causing virtual execution time to increase more slowly than a user task group with a lower fourth weight.
delete delete The method of claim 4, wherein the CPU resource scheduling method selects and schedules a process having a lowest virtual execution time and a user task group. 6. The method of claim 9, wherein the CPU resource scheduling method selects a tty task group having the lowest virtual execution time from the selected user task group, and has a lowest virtual execution time among the selected tty task groups. User-based resource scheduling method supporting a multi-user environment, characterized in that the process is selected. 12. The method of claim 10, wherein the CPU resource scheduling method uses all of the maximum virtual execution times that the selected process can execute or re-requests the CPU resource scheduling when the selected process is stopped. User-based resource scheduling method supporting multi-user environment. delete delete delete The method of claim 1, wherein the I / O resource scheduling method selects and schedules a user I / O group having the lowest virtual disk time, and the higher the weight, the slower the virtual disk time increases than the actual time used. Resource is scheduled so that the user I / O group having a first weight increases more time by causing the virtual disk time to increase more slowly than the user I / O group having a second weight lower than the first weight. Based resource scheduling method.
delete 2. The method of claim 1, wherein the input / output (I / O) resource scheduling method is a user input / output group having the lowest virtual disk time is selected and scheduled.

delete delete 18. The method of claim 17, wherein the method for scheduling input / output (I / O) resources schedules a user input / output group having the lowest virtual disk time to be selected between the user input / output groups, and a process or a process input / output queue in the user input / output group is round-robin User-based resource scheduling method for supporting a multi-user environment, characterized in that the scheduling in a manner.

Images