KR101444655B1 - Recording medium recording extended tmo model for partition computing, and implementation method of two step scheduling for extended tmo model and computer-readable recording medium recording the method - Google Patents

Abstract

The present invention relates to a recording medium which stores a TMO extension model for partition computing, a method for performing two step scheduling of the TMO extension model, and a computer readable recording medium which records the method. The recording medium which stores the TMO extension model for the partition computing performs the partition computing by adding a time condition to a time-triggered message-triggered object (TMO) model. The time condition comprises an activation period for each object which defines one partition; initial offset; and operation duration which processes activation every activation period.

Description

TECHNICAL FIELD [0001] The present invention relates to a recording medium storing a TMO extended model for partition computing, a method for implementing a two-step scheduling of a TMO extended model, and a computer readable recording medium recording the method. STEP SCHEDULING FOR EXTENDED TMO MODEL AND COMPUTER-READABLE RECORDING MEDIUM RECORDING THE METHOD}

The present invention relates to a recording medium storing a TMO extended model for partitioned computing that extends a distributed real-time object model TMO (Time-triggered Message-triggered Object) and a method for implementing a two- To a recording medium on which the recording medium can be recorded.

Time-triggered Message-triggered Object (TMO), which is a distributed real-time object model, is an object-independent and persistent object data store (ODS) in an object, a time- triggered spontaneous method SpM) and Message-triggered Server Method (SvM).

On the other hand, ARINC653, a standard of aircraft software architecture for modular architecture, defines a partition model in which CPU time and memory resources are fixedly divided on one hardware platform and used independently without interfering with each other.

Partitions are independent applications for controlling each part of an aircraft, and are a model for independently performing cabling in a single multiprocessor system, which is composed of a conventional federated system and a distirbuted system.

SUMMARY OF THE INVENTION The present invention has been made in view of the above problems, and it is an object of the present invention to provide a method for implementing a two-stage scheduling of a TMO extended model for partitioned computing, And a recording medium.

The storage medium storing the TMO extension model for partition computing according to an embodiment of the present invention can perform partition computing by adding a time condition to a time-triggered message-triggered object (TMO) model, The time condition may include an activation period for each object defining a partition, an initial offset that is a first activation time, and an execution duration in which activation is performed for each activation period .

The method of implementing the two-stage scheduling of the TMO extended model for partition computing according to an embodiment of the present invention includes a TMO extended model capable of performing partition computing independently using a CPU and a memory, triggered Message-triggered Object partition model: TMO.p model), wherein TMO.p means an object defining a partition, and (a) an activation cycle for each of a plurality of TMO.p receiving an initial offset which is a first activation time and an execution duration during which activation is performed for each activation period; (b) inputting TMO.p, to which a plurality of time-triggered spontaneous method (SPM) and message-triggered server method (SvM) belong, respectively; (c) a first partition scheduling step of activating one TMO.p having an activation period out of the plurality of TMO.p; And (d) a second task scheduling step of activating SpM and SvM belonging to the activated one TMO.p among the plurality of SpM and SvM.

In an embodiment, the step (b) may include: determining a scheduling policy for each of the plurality of SpMs, a deadline for the execution period, a deadline for each of the plurality of SvMs, and a plurality of SpMs and SvMs And receiving an input.

In an exemplary embodiment, the scheduling policy may include an Earliest Deadline First (EDF) scheme and a Least Laxity First (LLF) scheme in which preemption is permitted and a First Triggered First Scheduled (FTFS) scheme in which a preemption is not allowed.

In the embodiment, (c) the first partition scheduling step may include: if the second TMO.p is active before activating the first TMO.p when the activation period arrives, the execution period of the second TMO.p is And turning the second TMO.p into an inactive state when the first TMO.p is terminated, and activating the first TMO.p.

The method may further include: (d) switching the active SpM and SvM belonging to the TMO.p to an inactive state when the execution period of one active TMO.p is terminated.

A computer-readable recording medium recording a method for implementing a two-step scheduling of a TMO extension model for partition computing according to an embodiment of the present invention includes partition computing in which a CPU and a memory are separated and used independently The TMO.p is an object defining one partition. The TMO.p is an object defining a partition, and the TMO.p is an object defining a partition. (a) receiving an activation period for each of a plurality of TMO.p, an initial offset that is a first activation time, and an execution duration in which activation is performed for each activation period; (b) inputting TMO.p, to which a plurality of time-triggered spontaneous method (SPM) and message-triggered server method (SvM) belong, respectively; (c) a first partition scheduling step of activating one TMO.p having an activation period out of the plurality of TMO.p; And (d) a second task scheduling step of activating SpM and SvM belonging to the activated TMO.p among the plurality of SpM and SvM. 2. The method according to claim 1, Step scheduling implementation method.

In an embodiment, the step (b) may include: determining a scheduling policy for each of the plurality of SpMs, a deadline for the execution period, a deadline for each of the plurality of SvMs, and a plurality of SpMs and SvMs And receiving an input.

In an exemplary embodiment, the scheduling policy may include an Earliest Deadline First (EDF) scheme and a Least Laxity First (LLF) scheme in which preemption is permitted and a First Triggered First Scheduled (FTFS) scheme in which a preemption is not allowed.

In the embodiment, (c) the first partition scheduling step may include: if the second TMO.p is active before activating the first TMO.p when the activation period arrives, the execution period of the second TMO.p is And turning the second TMO.p into an inactive state when the first TMO.p is terminated, and activating the first TMO.p.

In the embodiment, (e) when the execution period of one activated TMO.p is terminated, switching SpM and SvM that are active in the TMO.p to the inactive state may be further included.

According to the present invention, kernel-level partition scheduling can be provided by introducing the concept of partitioned computing while maintaining the formal advantages of time-driven and event-driven member task support that can provide a deadline of the existing TMO model .

In the TMO.p model according to the present invention, as a member task of the TMO.p object, the process type in the existing ARINC653 specification distinguishes whether it is a periodic / non-periodic process and whether it is a deadline, And a non-periodic task, which is driven by a deadline based on the reception of an event message generated from the inside and the outside of the partition, and has the following advantages.

First, it is easy to design according to the characteristics of real-time control application such as time and event-based driving. Second, since TMO.p is an object that is a member of dynamic tasks, independent parallel programming It is possible to use object-oriented programming using inheritance, and it is possible to use scheduling analysis tool, design and automatic code generation tool developed for existing TMO system.

Also, because API for TMO.p is based on API of existing TMO engine, it does not exactly match partition API of ARINC653, but its functional aspects are almost same. Therefore, it is difficult to make aircraft software by partition-based TMO.p It can be said that there is no.

1 is a conceptual diagram for explaining the structure of a TMO-based partition model TMO.P.
FIG. 2 is a conceptual diagram for explaining a Watchdog Real-time Scheduler for TMO.p, which is a two-stage scheduler composed of TMO.p partitions and respective schedulers for each partition belonging task.
FIG. 3 is a conceptual diagram showing a state transition of a task by the first-stage partition scheduler and the second-stage task scheduler.
4 is a conceptual diagram for explaining an implementation of the two-stage scheduler in a clock interrupt handler.
5 is a conceptual diagram showing an embodiment of calculation of a non-overlapping partition scheduling scenario.

Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings so that those skilled in the art can easily carry out the technical idea of the present invention. The present invention may, however, be embodied in many different forms and should not be construed as limited to the embodiments set forth herein. In order to clearly explain the present invention, parts not related to the description are omitted, and like parts are denoted by similar reference numerals throughout the specification.

The TMO extended model TMO.p (TMO.partition) for partition computing according to the present invention is a TMO (Partitioned TMO) partitioning method for partition computing in which a time-triggered message- triggered Object) model.

This TMO.p model has the following differences and common points in comparison with the existing TMO model.

In the TMO.p model according to the present invention, the existing TMO model only presents the scheduling conditions for the member tasks in the object, but the TMO.p model according to the present invention, It should specify the consecutive CPU occupation time, memory requirement, port for communication between partitions.

Common features include a set of concurrent tasks, a task configuration consisting of periodic or sporadic tasks, a deadline imposed, a channel-based communication tool, And the data storage of the object is independent and static in the object.

Therefore, when the TMO is executed with deterministic scheduling without overlapping independently on a single SMP (Symmetric Multi-processor) by giving the CPU occupation time during periodic and periodic execution, the concept of partitioned computing Can be introduced.

The TMO.p, which means TMO.partition, is named as TMO.p and TMO.p is a model that specifies the period of the CPU and the period of each cycle and the port of the communication channel. In this specification, it is described in synonym with TMO extension model for partitioned computing do.

FIG. 1 is a conceptual diagram for explaining the structure of a TMO-based partition model TMO.p.

Referring to FIG. 1, TMO.p is an object defining one partition, and has a period of activation, an initial offset which is the first activation time, and an execution duration of each activation Time condition.

On the other hand, TMO.p has a time-triggered spontaneous method (SpM), which is driven by a time condition given the execution cycle and the deadline as in the existing TMO, (Message-triggered Server Method: SvM).

Unlike in the TMO, however, these tasks are only activated when the TMO.p partition reaches its execution boundary. Since multiple TMO.p partitions use separate CPU time without nesting and only use ODS (Object Data Store) which is limited to objects in memory, it is necessary to satisfy the condition of partition to use independent CPU / memory, It has the advantage of real-time task support.

FIG. 2 is a conceptual diagram for explaining a Watchdog Real-time Scheduler for TMO.p, which is a two-stage scheduler composed of TMO.p partitions and respective schedulers for each partition belonging task.

Referring to FIG. 2, after the TMO.p time condition registration is completed, each SpM and SvM tasks must also register the time condition of the task, the scheduling policy, and the belonging TMO.p in the kernel. In this case, it is necessary to define the time flow for processing the deadline given to SpM and deadline and SvM, and the time flow for TMO.p and its affiliated task model is defined as dependent on each TMO.p do.

That is, the time of SpM and SvM elapses only when the belonging TMO.p is in the activated state, and stops when the TMO.p is in the inactive state. When TMO.p is active and the execution interval has elapsed, the belonging SpM and SvM are immediately suspended, and their execution proceeds to the next cycle of the belonging TMO.p.

Since it is not desirable to stop the execution of the belonging SpM or SvM during execution in one section of TMO.p, it is common for the periodic SpM to be set to be synchronized with the period and execution period of TMO.p to be.

The scheduling policies of the member tasks have been developed for each task by developing the Earliest Deadline First (EDF), Least Laxity First (LLF), and First Triggered First Scheduled (FTFS) methods that do not allow preemption.

The scheduler for TMO.p of the present invention is WRST (Watchdog Real-time Scheduler for TMO.p), which is a two-step scheduler that adds partition scheduling to the task scheduler.

The two-stage scheduler consists of a first-stage partition scheduler in FIFO mode and a second-stage task scheduler in which EDF, LLF, and FTFS methods can be selected for each task.

The first stage partition scheduler activates the TMO.p with a period of time and activates the tasks of the belonging tasks by the task scheduler at the task level. If there is overlap of partition execution, activation of the subsequent partition is postponed until the end of the execution interval of the preceding partition. For the active TMO.p, the end of the execution period is checked to switch to the deactivation state to stop the time flow of the corresponding TMO.p.

The second stage task scheduler manages the task according to the ACTIVE / INACTIVE state of the TMO.p belonging to the task. The main states of the SpM and SvM member tasks are shown in Table 1 below.

Status of SpM / SvM Explanation ACTIVE (SpM) of task cycle or event message reception (SvM) within the execution interval of TMO.p belonging to SpM or SvM. This state includes not only the actual execution state but also the standby state due to I / O or the like during execution. ACTIVE_
OUTSIDE_
DURATION When the execution interval of TMO.p expires while SpM or SvM is running, the execution and the time flow are stopped. These states are provided and handled, but it is best avoided by pre-synchronous design. INACTIVE The belonging TMO.p is in the execution period but is in a waiting state due to non-arrival of the period or non-arrival of the (SpM) message (SvM). INACTIVE_
OUTSIDE_
DURATION In the INACTIVE state, there is no modification to the dynamic time condition, such as the time remaining after the execution period of TMO.p is terminated and the time remaining until the task cycle.

Step 2 The task scheduler performs the following tasks on the SpM or SvM in ACTIVE state.

Specifically, when the execution period of the belonging TMO.p is ended, execution of the SpM or SvM in the ACTIVE state is stopped and the state is changed to the ACTIVE_OUTSIDE_DURATION state.

On the other hand, if the execution period of the belonging TMO.p has not ended, the priority order of the SpM or SvM in the ACTIVE state is adjusted according to the scheduling policy given to each task, and a context exchange request is set.

In this case, the real-time scheduling policies provided include Earliest Deadline First (EDF), Least Laxity First (LLF), and First Triggered First Scheduled (FTFS) .

In addition, the tasks performed by the second-stage task scheduler on the SpM or SvM in the ACTIVE_OUTSIDE_DURATION state are as follows.

Specifically, when the periodical execution period of the belonging TMO.p is started, the belonging tasks are switched to the ACTIVE state so that deadline-based scheduling can be received. On the other hand, before the arrival of the execution interval, the dynamic time condition modification operation is not performed according to the time flow of the task.

The tasks performed by the Task Scheduler in the INACTIVE state for SpM and SvM are as follows.

Specifically, when the execution period of the belonging TMO.p is ended, the stop count (suspend_count) of the SpM or SvM in the INACTIVE state is increased to stop in the overlap waiting state, and the state is changed to the INACTIVE_OUTSIDE_DURATION state. In this case, the stop count becomes 2, and the state is switched to the ACTIVE state only when both the original waiting condition (arrival of the task cycle and arrival of the message) and arrival of the execution interval are satisfied.

On the other hand, if it is not the end of the execution period, it checks the arrival of the task cycle of the SpM and determines whether or not to switch ACTIVE. In the case of SvM, the IPC system switches to ACTIVE upon arrival of a message, so the scheduler does not perform any operation.

In addition, the tasks performed by the second-stage task scheduler on the SpM or SvM in the INACTIVE_OUTSIDE_DURATION state are as follows.

Specifically, if the periodical execution period of the belonging TMO.p has arrived, the stop count of the belonging tasks is reduced, and the state is switched to the INACTIVE state so that it can be driven by the original task time condition or message delivery.

On the other hand, before the arrival of the execution interval, the dynamic time condition modification operation is not performed according to the time flow of the task.

FIG. 3 is a conceptual diagram showing a state transition of a task by the first-stage partition scheduler and the second-stage task scheduler.

Referring to FIG. 3, a partition belonging SpM is changed to an active state when a task cycle arrives when a partition is activated, and to an INACTIVE state when a task is periodically performed.

However, when the partition becomes inactive, it will stop in ACTIVE_OUTSIDE_DURATION state when the cycle of the task arrives, and will transition to ACTIVE state when the partition is activated.

When the partition is activated, the SpM that was in the INACTIVE state after the periodic execution is double-suspended to the INACTIVE_OUTSIDE_DURATION state when the partition is disabled. Such an SPM is switched to the INACTIVE state when the partition is activated.

Partition belongs to SvM. When partition is activated, it goes into ACTIVE state when message arrives, and to INACTIVE state when execution of message processing is completed.

However, if a message arrives when the partition is inactive, it will stop in ACTIVE_OUTSIDE_DURATION state and transition to ACTIVE state when the partition is activated.

Likewise, SvM in the INACTIVE state when the partition has been activated when the message processing is finished, will be suspended in the INACTIVE_OUTSIDE_DURATION state when the partition is disabled. These SvMs are switched to the INACTIVE state when the partition is activated.

4 is a conceptual diagram for explaining an implementation of the two-stage scheduler in a clock interrupt handler.

Referring to FIG. 4, the two-stage real-time scheduler is implemented in the clock interrupt handler of the RT-eCos. The clock interrupt handler can be divided into an Interrupt Service Routine (ISR) and a Deferred Service Routine (DSR).

Specifically, the Interrupt Service Routine (ISR) operates regardless of whether a scheduler at the time of interruption or a system call is performed and can perform only a basic operation such as a clock counter reset. DSR (Deferred Service Routine) delays the process until the scheduler lock is released when scheduler lock is set by scheduler or system call.

Since the two-stage real-time scheduler according to the present invention requires mutual exclusion from other scheduling actions in the kernel, it can be implemented as shown in FIG. 4 in the DSR.

Specifically, the DSR is performed immediately after the ISR in the tick or at the end of the kernel's external interrupt processing, at the end of the kernel critical section, and so on. In the DSR, the first and second WRST real time schedulers are run in addition to the routine tasks such as the modification of the system time, the timer related service, and the modification of the remaining time slice.

On the other hand, in order to avoid overlapping of the periodic execution of each TMO.p, the initial offset must be set so that there is no overlapping of execution of TMO.p within a hyper-period which is the least common multiple of the cycles. It is called a fire problem.

In order to serialize the partitions, a task serializer can be used for sequencing the periodical task. The task sequencer is a tool that determines the initial phase by using a predefined execution interval and cycles so that there is no overlap of periodic partition execution. In the case where there is no output of this tool, it is impossible to have a scenario without overlap. If nesting of execution is inevitable, execution of the nested partitions is deferred until completion of the preceding partition.

5 is a conceptual diagram showing an embodiment of calculation of a non-overlapping partition scheduling scenario.

Referring to FIG. 5, it can be seen that each cycle and execution intervals of the nine TMOs are input to obtain an initial phase without overlapping while satisfying an accurate periodic execution start time of the partitions.

Here, for convenience hyper period of 6 T interval (T being the least common multiple of the partition period) it is displayed by each divided into six slots of a width T.

For example, in the case of TMO.p ₂ in FIG. 5, 2 e (where e is a tick of 10 milli-sec of RT-eCos as the unit execution time, T = 5 e in this embodiment) Phase, so that the execution can be repeated every 3 T cycles without overlap with other partitions.

As a result, according to the present invention, kernel-level partition scheduling is provided by introducing the concept of partitioned computing while maintaining the formal advantages of time-driven and event-driven member task support capable of providing deadlines of existing TMO models can do.

In the TMO.p model according to the present invention, as a member task of the TMO.p object, the process type in the existing ARINC653 specification distinguishes whether it is a periodic / non-periodic process and whether it is a deadline, And a non-periodic task, which is driven by a deadline based on the reception of an event message generated from the inside and the outside of the partition, and has the following advantages.

First, it is easy to design according to the characteristics of real-time control application such as time and event-based driving. Second, since TMO.p is an object that is a member of dynamic tasks, independent parallel programming It is possible to use object-oriented programming using inheritance, and it is possible to use scheduling analysis tool, design and automatic code generation tool developed for existing TMO system.

Also, because API for TMO.p is based on API of existing TMO engine, it does not exactly match partition API of ARINC653, but its functional aspects are almost same. Therefore, it is difficult to make aircraft software by partition-based TMO.p It can be said that there is no.

As described above, the method of implementing the TMO extension model for partition computing is not limited to the configuration and method of the embodiments described above, but the embodiments may be modified so that all or some of the embodiments may be modified Or may be selectively combined.

Claims (11)

Partition computing can be performed by adding a time condition to a time-triggered message-triggered object (TMO) model,
The time conditions include,
Wherein the activation period includes an activation period for one partition, an initial offset that is a first activation time, and an execution duration during which activation is performed for each activation period. (Time-triggered Message-triggered Object Partition Model: TMO.p model). TMO.p means one partition,
(a) receiving an activation period for each of a plurality of TMO.p, an initial offset which is a first activation time, and an execution duration in which activation is performed for each activation period, ;
(b) receiving a TMO.p in which a second input unit belongs to each of a plurality of time-triggered spontaneous methods (SpM) and a message-triggered server method (SvM);
(c) a first stage partition scheduler activating one TMO.p with an activation period of the plurality of TMO.p; And
(d) activating the SpM and SvM belonging to the activated one TMO.p among the plurality of SpM and SvM by the second-step task scheduler, A two-step scheduling implementation method of a TMO extended model (TMO.p model). 3. The method of claim 2,
The step (b)
And a second input unit receiving an execution cycle and a deadline for each of the plurality of SpMs, a deadline for each of the plurality of SvMs, and a scheduling policy for each of the plurality of SpMs and SvMs A Two - Step Scheduling Implementation Method of TMO Extended Model for Partitioned Computing. The method of claim 3,
Wherein the scheduling policy includes an EDF (Earliest Deadline First) and an LLF (Least Laxity First) method in which preemption is allowed and a FTFS (First Triggered First Scheduled) method in which preemption is not allowed. A method for implementing the two-step scheduling. 3. The method of claim 2,
The step (c)
If the second TMO.p is active before activating the first TMO.p that has arrived at the activation period,
And a step of activating the first TMO.p after the first-stage partition scheduler switches the second TMO.p to an inactive state when the execution period of the second TMO.p is terminated A Two - Step Scheduling Implementation Method of TMO Extended Model for Partitioned Computing. 3. The method of claim 2,
(e) when the task scheduler of the second stage finishes the execution period of one activated TMO.p, it turns on the active SpM and SvM belonging to TMO.p to the inactive state Two - Step Scheduling Implementation of TMO Extended Model for Partitioned Computing. TMO.p means one partition,
(a) receiving an activation period for each of a plurality of TMO.p, an initial offset which is a first activation time, and an execution duration in which activation is performed for each activation period, ;
(b) receiving a TMO.p in which a second input unit belongs to each of a plurality of time-triggered spontaneous methods (SpM) and a message-triggered server method (SvM);
(c) a first stage partition scheduler activating one TMO.p with an activation period of the plurality of TMO.p; And
(d) a second task scheduler activating a SpM and SvM belonging to the activated one TMO.p among the plurality of SpM and SvM; and a second task scheduling step A computer-readable recording medium recorded with a program for causing a computer to execute a two-step scheduling implementation method of a TMO extended model (TMO.p model). 8. The method of claim 7,
The step (b)
And a second input unit receiving an execution cycle and a deadline for each of the plurality of SpMs, a deadline for each of the plurality of SvMs, and a scheduling policy for each of the plurality of SpMs and SvMs A computer readable recording medium having a program for executing a two-step scheduling implementation method of a TMO extended model for partitioned computing. 9. The method of claim 8,
Wherein the scheduling policy includes an EDF (Earliest Deadline First) and an LLF (Least Laxity First) method in which preemption is allowed and a FTFS (First Triggered First Scheduled) method in which preemption is not allowed. And a program for causing a computer to execute the two-step scheduling implementation method. 8. The method of claim 7,
The step (c)
If the second TMO.p is active before activating the first TMO.p that has arrived at the activation period,
And a step of activating the first TMO.p after the first-stage partition scheduler switches the second TMO.p to an inactive state when the execution period of the second TMO.p is terminated A computer readable recording medium having a program for executing a two - step scheduling implementation method of a TMO extended model for partition computing. 8. The method of claim 7,
(e) when the task scheduler of the second stage finishes the execution period of one activated TMO.p, it turns on the active SpM and SvM belonging to TMO.p to the inactive state A computer readable recording medium having a program for executing a two - step scheduling implementation method of a TMO extended model for partitioned computing.

Images