METHOD FOR COMPOSING AND EXECUTING A REAL-TIME TASK SEQUENCE PLAN

Abstract

A method for executing two tasks in timesharing, includes: decomposing offline each task in a repetitive sequence of consecutive frames, and defining a start date and deadline by which an associated atomic operation must respectively start and end; verifying for each frame of a first of the repetitive sequences the corresponding operation can be performed between any two successive operations of a group of frames of the second repetitive sequence, overlapping the first repetitive sequence frame; and if the verification is satisfied, allowing the execution of the two tasks. Scheduling the operations of the two tasks, if two operations can start, executing the operation having the shorter deadline; and if a single operation can start, executing it if its execution need is less than the time remaining until the next frame start date of the other sequence, plus the time margin associated with the next frame of the other sequence.

Description

FIELD

The invention relates to multi-tasking real-time critical systems, such as control and monitoring systems used in the fields of transport (automotive, railway, naval, aerospace), industrial automation, energy, as well as other systems for which controlled allocation of resources is important, as in communications networks. The invention relates both to real-time systems for which a high level of implementation performance is required and to critical systems for which a high level of warranty is required as regards the use of hardware resources allocated to each task in the system.

BACKGROUND

In critical real-time systems, i.e. systems that do not tolerate any deadline violation through a belated execution of an operation, tasks are often executed through static scheduling methods. A static temporal allocation of the need for execution resources is then constructed offline, which demonstrates the temporal independence of tasks between them as regards the use of resources, and in particular the processor. This approach is described, for example, in the article [“A method and a technique to model and ensure timeliness in safety critical real-time systems”, C. Aussaguès, V. David, Fourth IEEE International Conference on Engineering of Complex Computer Systems, 1998], and in patent applications WO2006-050967 and US2010-0199280.

However, this approach requires considering only a single sequencing plan for all activities, thus considering only a single time base for determining deadlines. This makes implementations without interference on a single processor difficult, for tasks with uncorrelated time bases, such as a task using a clock to determine its deadlines and a task using a position of a variable speed object to determine its deadlines. This would imply building two sequencing plans clocked by a common time base so that they can be composed in a single sequencing plan without interference between tasks. This is achievable in practice only where the resources are largely oversized to account for the worst-case scenario of each situation.

In some real-time systems having a high level of performance, that are not defined as “critical” because they can tolerate deadline violations within a certain margin, the composition of two sequencing plans clocked by different time bases is performed by dynamic scheduling algorithms. Such algorithms are described, for example, in [“Scheduling algorithms for multiprogramming in a hard real-time environment”, C. Liu, J. Layland, Journal of the ACM, vol. 20, no. 1, pp. 46-61] and [“Foundations of Real-Time Computing: Scheduling and Resource Management”, edited by André M. Van Tilborg, Gary M. Koob, 1991, Kluwer Academic Publishers] and also in [“A method and a technique to model and ensure timeliness in safety critical real-time systems”, C. Aussaguès, V. David, Fourth IEEE International Conference on Engineering of Complex Computer Systems, 1998].

With these algorithms, preemptions are inevitable, i.e. the operating system can interrupt an ongoing operation to execute a more urgent operation. Scheduling being dynamic, the number and duration of the preemptions is not determinable, and each preemption introduces an overhead to perform context switching. This causes interference between tasks, so that the systems are more difficult to size, less predictable and not reproducible on complex processor architectures. Designers of such systems also encounter difficulties to properly set task priorities or deadlines in a system with two different time bases. The chosen options, such as the variation in execution needs depending on the values of application parameters, make systems constructed in this way very complex and uncontrollable.

SUMMARY

There is thus a need for a technique capable of executing in time sharing two sequencing plans without interference, even if the time bases of the two plans are different.

This need is addressed by providing a method for executing two tasks in timesharing, comprising the steps of decomposing offline each task in a repetitive sequence of consecutive frames in a time base associated with the task, wherein each frame is associated with an atomic operation having an execution need, and defines a start date from which the operation may start and a deadline by which the operation must end, whereby each frame defines a time margin in which the operation may start; verifying for each frame of a first of the repetitive sequences that the corresponding operation can be performed between any two successive operations of a group of frames of the second repetitive sequence, overlapping the frame of the first repetitive sequence, while respecting the start dates and deadlines of the operations; and if the verification is satisfied, allowing the execution of the two tasks.

The operations of the two tasks are then scheduled according to the following cases, evaluated at each key date, wherein a key date occurs at each deadline, each start date, and each end of operation: if two operations can start, executing the operation having the shorter deadline; and if a single operation can start, executing it only if its execution need is less than the time remaining until the start date of the next frame of the other sequence, plus the margin of time associated with said next frame of the other sequence.

Each frame may further comprise a reserve time interval that may be zero, located after the deadline of the frame or before the start date of the frame, wherein the reserve time intervals of any one of the repetitive sequences may be used to execute an operation of the other repetitive sequence.

The verification step may comprise the following steps:

• • for each frame Fa_i of the first repetitive sequence, evaluating the following conditions for each frame Fb_j of the second repetitive sequence:

Da _i ≦Tb _j+1  [A.1]

Da _i ≦Rb _j +Tb _j+1  [B.1]

and Db_j≦Ta_i, and

Db _j ≦Ta _i,  [C.1]

• • where D, T and R are the execution need, the time margin, and reserve time, indexed by a repetitive sequence identifier and the frame number in the sequence;
  • repeating the condition evaluations with the frames Fa_i and Fb_j exchanged;
  • authorizing execution of the tasks if the three conditions [A.1], [B.1] and [C.1] are satisfied for any pair (i, j).

The verification step may comprise the following additional steps:

• • if the condition [A.1] is false, evaluating the following condition:

Da _i ≦Tb _j+1 +Rb _j+1 +Tb _j−2  [A.2]

and Da_i+Db_j+1≦Ta_i+Tb_j−1

• • if the condition [B.1] is false, evaluating the following condition:

Da _i ≦Tb _j+1 +Rb _j+1 +Tb _j+2  [B.2]

and Db_j+Da_i+Db_j+1≦Ta_i+Tb_j+1

• • if the condition [C.1] is false, evaluating the following condition:

Db _j ≦Ta _i+min[Tb_j−Da_i,Ra_i,Ra_i−1]  [C.2]

• • authorizing execution of the tasks if the conditions [A.1] or [A.2] and [B.1] or [B.2] and [C.1] or [C.2] are met for each pair (i, j).

The time bases determining the durations of the frames may be variable so that the time margins T and the reserve times R vary by a factor 1 to za for the frames Fa and a factor 1 to zb for the frames Fb. The verification step may then further comprise the steps of:

• • if the condition [A.1] is false, evaluating the following condition:

Da _i ≦Tb _j+1 +Rb _j+1 +Tb _j+2  [A.2]

and min[ql(zb·(Tb_j+Rb_j)+(zb−1)Db_j),Da_i]+Db_j+1≦Ta_i

• • where ql=(Da_i−Tb_j+i)/(Tb_j+Rb_j);
  • if the condition [B.1] is false, evaluating the following condition:

Da _i ≦Tb _j+1 +Rb _j+1 +Tb _j+2  [B.2]

and Db_j+min[ql(zb·Tb_j+(zb−1)Db_j)+zb·Rb_j,Da_i]+Db_j+1≦Ta_i

• • where ql=(Da_i−Rb_j−Tb_j+1)/Tb_j;
  • if the condition [C.1] is false, evaluating the following condition:

Db _j ≦Ta _i+min[Tb_j−Da_i,Ra_i,Ra_i−1]  [C.2]

• • authorizing execution of the tasks if the conditions [A.1] or [A.2] and [B.1] or [B.2] and [C.1] or [C.2] are met for each pair (i, j).

The verification step may comprise the following steps:

• • if the condition [B.1] is false, evaluating the following condition before the condition [B.2]:

Da _i ≦Tb _j +Rb _j +Tb _j+1  [B.1.1]

and Db_j−1≦Tl+min[Tb_j−1+Db_j−1+Rb_j−1−Tl−Da_i−1,Ra_i−2]

• • where ql=(Da_i−Rb_j−Tb_j+1)/Tb_j and Tl=Ta_i−1−(1−ql)Tb_j+Ra_i−1
  • authorizing execution of the tasks if the conditions [A.1] or [A.2] and [B.1] or [B.1.1] and [C.1] or [C.2] are met for each pair (i, j).

BRIEF DESCRIPTION OF DRAWINGS

Other advantages and features will become more clearly apparent from the following description of particular embodiments of the invention provided for exemplary purposes only and represented in the appended drawings, in which:

FIG. 1 shows an example of a repetitive sequence of frames used for composing a sequencing plan for a task;

FIG. 2 shows the structure of a frame of a repetitive sequence and different possibilities of starting an operation in the frame;

FIGS. 3A to 3C illustrate different possibilities of time-sharing two repetitive sequences of frames, where FIG. 3C illustrates an error condition;

FIGS. 4A to 4C illustrate time-sharing situations of two sequences according to a first assumption for establishing validation criteria of a composition of two sequences;

FIGS. 5A and 5B illustrate time-sharing situations of two sequences according to a second assumption for establishing validation criteria of a composition of two sequences;

FIG. 6 illustrates a time-sharing situation of two sequences according to a third assumption for establishing validation criteria of a composition of two sequences;

FIG. 7 illustrates an example of a succession of frames for establishing another validity condition in the situation of FIG. 5A; and

FIG. 8 shows an additional execution situation of two sequences according to the third assumption, allowing to establish alternative validity conditions.

DESCRIPTION OF EMBODIMENTS

A real-time task may be considered as a sequence of basic processes or operations to be performed one after the other, where each operation has a deadline in a time base associated with the task. In a critical real-time system, deadlines are mandatory, i.e. an operation that exceeds its deadline would cause an error condition and a system shut-down, for example. The techniques described below are applicable to such a system to improve the level of performance in that multiple tasks, even with uncorrelated time bases, may be executed in time-sharing on the system.

FIG. 1 illustrates a decomposition of a task into a repetitive sequence of frames RSF, where each frame F is assigned to a sequential operation of the task. The task is implemented by executing the repetitive sequence RSF in a loop.

FIG. 2 shows a temporal structure of a frame F. The start of the frame is the date from which the operation may begin. The frame defines a deadline E at which the operation should be completed. The operation is assumed to be atomic, i.e. the operation cannot be suspended to perform another operation. The deadline E may precede the end of the frame by an optional reserve time slot R that may be zero. A time interval D is an execution need of the operation associated with the frame. The execution need D corresponds, for example, to the number of processor clock cycles required to execute the operation—it may thus be expressed as a fixed time depending on the characteristics of the processor of the target system. An interval T corresponds to a time margin between the start of the frame and the deadline E to execute the operation.

Thus, as shown in FIG. 2, the operation may begin anywhere between the start of the frame, where the interval T lies between the end of the operation and the deadline, and a time T after the start of the frame, where the operation ends just at the deadline E. For intermediate situations, qT denotes the interval between the end of the operation and the deadline, with 0≦q≦1, and (1−q)T denotes the complementary interval between the beginning of the frame and the beginning of the operation.

The sum of the reserve intervals R is, for example, a time that the programmer allocates the system to conduct non real-time operations, such as input/output.

Two tasks that should be performed in time-sharing on critical real-time system are thus broken down into repetitive sequences of frames. For the tasks to be executable without interference, i.e. respecting the atomicity of operations, it is sought to compose the sequences so that each operation of one sequence can be executed at any time between two successive operations of the other sequence, while respecting the constraints of the frames of the two sequences, namely the deadlines and frame starts.

FIGS. 3A-3C show various situations that may be encountered while unfolding two sequences RSFa and RSFb. It is assumed, for sake of simplicity, that the reserve intervals R are zero.

In FIG. 3A, as an example, the sequence RSFa includes three frames Fa₁ to Fa₃ of respective parameters (Da₁=3, Ta₁=6), (Da₂=4, Ta₂=1), and (Da₃=1, Ta₃=1), where time values are expressed in normalized units. The sequence RSFb begins 3 units after the start of frame Fa₁ and it comprises two frames Fb₁ and Fb₂ of respective parameters (Db₁=4, Tb₁=5) and (Db₂=3, Tb₂=4).

The operation Da₁ begins one unit after the start of the corresponding frame. At the end of operation Da₁, the operation Da₂ cannot be started yet, but operation Db₁ may be started. At the end of operation Db₁, no operation is ready for execution. The start of frame Fa₂, here deadline Ea₁, is awaited for starting the operation Da₂. At the end of the operation Da₂, the operation Db₂ is executable, but it should not be started. Indeed, if operation Db₂ were launched immediately, it would end just at the deadline Ea₃, whereby the operation Da₃ could not run before that deadline. In this situation, the beginning of the frame Fa₃ is awaited to launch the operation Da₃. Finally, the operation Db₂ can start at the end of operation Da₃.

In FIG. 3B, the same parameters are preserved, except that the operation Da₁ starts three units after the beginning of the frame. The operations Db₁ and Da₂ are delayed by two units with respect to FIG. 3A, while respecting the constraints of the corresponding frames. The operation Da₂ ends just at its deadline Ea₂.

In FIG. 3C, the same parameters are preserved, except that the operation Da₁ now starts four units after the beginning of the frame. The operations Db₁ and Da₂ are delayed by three units with respect to FIG. 3A. The operation Db₁ can still be executed while respecting its deadline Eb₁, but the operation Da₂ violates its deadline Ea₂.

A first step common to the techniques referred to herein includes the composition, for two tasks to execute in time-sharing, of two repetitive sequences RSFa and RSFb that can be unwound in the conditions of FIG. 3A or 3B, regardless of the delay between the beginnings of the sequences. This composition is static, that is to say achieved offline by the programmer.

A second step includes verifying that any two sequences are compatible, i.e. that they can be unwound so that each operation can be executed within its constraints, such as in FIGS. 3A and 3B.

This compatibility verification may be implemented in a development tool designed for the programmer to compose sequences by trial and error. It is preferably also carried out in the real-time system itself, upon loading for execution sequencing plans defined by the two repetitive sequences. The system may then be configured to refuse operation if the sequences are detected as incompatible.

A third step includes scheduling the operations of the two sequences in the real-time system. As the figures show, it is not sufficient for achieving correct operation to alternate the operations of the first sequence with the operations of the second sequence or to start an operation when it is ready for execution. Scheduling may be achieved according to the following cases, assessed at each key date. Key dates may be defined as the beginnings of the frames in the two sequences, the deadlines, and the operation ends. When the reserve time slots R (FIG. 2) are zero, like in the presented examples, deadlines coincide with frame starts. For a given key date associated with a first of the sequences, the term “concurrent frame” denotes the current frame of the second sequence.

• • 1) No operation can start. This occurs when an operation is already in progress (as at deadline Eb1 of FIG. 3A), or at the end of an operation that runs between the operation of the concurrent frame and its deadline, as for operation Db1 of FIG. 3A. In this case, the next key date is awaited.
  • 2) Two operations can start. This usually happens when starting a frame while an operation is waiting in the concurrent frame, as is the case at deadline Ea2 of FIGS. 3A and 3B. The operation having the shortest deadline is then executed (Da3 in FIGS. 3A and 3B). The case also occurs when two concurrent frames start at the same time.
  • 3) A single operation can start. This usually happens at the end of an operation (Da1), or at the beginning of a frame when the operation of the concurrent frame has ended (Ea1, FIG. 3A). The operation is executed only if its execution need is less than, or equal to the time remaining until the start of the next concurrent frame, plus the time margin associated with this concurrent frame. For example, in FIG. 3A, the operation Da2 is ready to start at deadline Ea1, there are 3 units left until the beginning of frame Fb2 and Tb2=4. The operation Da2 is started because Da2=4<3+4.

Otherwise, the next key date is awaited to reassess the situation. For example, in FIG. 3A, the operation Db2 is ready for execution at the end of the operation Da2. However, there is one unit left until the beginning of the frame Fa3 and Ta3=1, hence Db2=3>1+1.

Compatibility criteria will now be defined between any two repetitive sequences RSFa and RSFb used in sequencing plans for a real-time system.

An exhaustive compatibility check would involve examining all possible combinations of positions of the operations in their frames and all possible positions of the sequences relative to each other, with a granularity of one clock cycle, to assess whether concurrent operations can be executed according to the imposed constraints.

Such an exhaustive verification would be difficult to achieve in a reasonable time. Simplified criteria are sought that can be evaluated online on the real-time system upon loading sequencing plans accompanying executable code for the tasks. For this purpose, three general assumptions are considered regarding the positioning of each frame of one of the sequences with respect to each frame of the other sequence. For each assumption, it is verified that the operation of a given frame Fa_i can be executed within an available slot left by the concurrent frames Fb_j, Fb_j+, . . . overlapping the frame Fa_i. Two frames Fa_i and Fb_j are considered compatible when the verification is satisfied for each of the three assumptions.

• • Assumption (A): the given frame Fa_i starts when the operation Db_j of the first concurrent frame Fb_j has completed execution.
  • Assumption (B): the operation of the first concurrent frame is running.
  • Assumption (C): the operation of the concurrent frame has not started.

For each assumption, criteria are sought that are independent of the relative positions of the frames. For this purpose, in general, unknown time intervals used to find slots to perform operations are underestimated, and unknown time intervals used to check compliance with deadlines are overestimated.

FIGS. 4A to 4C illustrate exemplary frames that may be used to develop compatibility criteria based on assumption (A). The given frame Fa_i starts during the concurrent frame Fb_j, when the operation Db_j of this concurrent frame has completed. The frame Fa_i overlaps the next concurrent frame Fb_j+1. All that is known about the beginning of the frame Fa_i is that it is located in an interval qTb_j following the end of the operation Db_j, where 0≦q≦1.

In FIG. 4A, the frame Fa_i ends before the frame Fb_j+1. The operation Da_i could be launched from the end of the operation Db_j, in an interval qTb_j+Tb_j+1. But q being unknown and arbitrary, the worst-case scenario is considered with q=0, whereby the operation Da_i starts at the beginning of the frame Fb_j+1. Thus, the operation Da_i can be executed if:

Da _i <Tb _j+1.  [A.1]

In FIG. 4B, the operation Da_i does not meet the requirement A.1. It is then tried to execute the operation Da_i astraddle the frames Fb_j+1 and Fb_j+2, meaning that the frame Fa_i overlaps the three frames Fb_j to Fb_j+2. This implies satisfying both a condition to ensure that the sum of intervals Tb_j+1 and Tb_j+2 is sufficient, and a condition to ensure that the deadline Eai will be met:

Da _i ≦Tb _j+1 +Tb _j+2

AND

qTb _j +Db _j+1 +Da _i ≦Da _i +Ta _i, hence qTb_j+Db_j+1≦Ta_i

The worst case for respecting the deadline is q=1, increasing the left member of the inequality. Therefore the deadline Eai is met in all cases if:

Tb _j +Db _j+1 ≦Ta _i

However, it is not realistic to consider in a same context worst-case situations using the two extremes of a same variable. In fact, a certain range of values for the variable q can be excluded when considering the behavior of the real system in the situation of FIG. 4A.

FIG. 4C shows the behavior of the real system in the situation of FIG. 4A. Once the operation Db_j ends, the system, having on-line knowledge of the frame parameters, launches operation Da_i at the beginning of the frame Fa_i, as long as:

Da _i ≦qTb _j +Tb _j+1

that is to say, if conditions 3) of the scheduling method are met.

Let ql be the limit value of q to achieve equality in the above inequality:

Da _i =qlTb _j +Tb _j+1, hence ql=(Da_i−Tb_j+1)/Tb_j

If q≧ql, it is certain that the execution of the real system occurs according to FIG. 4C and that the situation of FIG. 4B is to be excluded. The worst case for FIG. 4B is then q=ql, whereby:

qlTb _j +Db _j+1 ≦Ta _i, hence, by replacing ql by its value,

Da _i −Tb _j+1 +Db _j+1 ≦Ta _i, or Da_i+Db_j+1≦Ta_i+Tb_j+1

The expression used to calculate the value ql is not limited to 1. If ql exceeds 1, it does not make practical sense. The value ql is thus capped at 1.

The conditions resulting from the situation of FIG. 4B are thus:

Da _i ≦Tb _j+1 +Tb _j+2

AND

qlTb _j +Db _j+1 ≦Ta _i  [A.2]

The condition A.1 is verifiable in a situation where the frame Fa_i overlaps at least two consecutive frames Fb. The condition A.2 is verifiable in a situation where the frame Fa_i overlaps at least three consecutive frames Fb. In a similar manner, conditions A.3 and following may be found that are verifiable in situations where the frame Fa_i overlaps at least four or more consecutive frames Fb, but the likelihood of satisfying such conditions decreases rapidly. The conditions A.1 and A.2 (and, if applicable, A.3 and following) are alternative, that is to say, it is sufficient that any one of these conditions is satisfied for jumping to the validation of the other assumptions.

FIGS. 5A and 5B illustrate examples of frames that may be used to develop compatibility criteria based on the assumption (B). The frame Fa_i starts during a concurrent frame Fb_j, while the concurrent operation Db_j of this frame is ongoing. In the worst case, as shown, the operation Db_j just starts. The situations are similar to those of FIGS. 4A and 4B, except that the duration Db_j is accounted for to verify compliance with the deadline Eai.

According to FIG. 5A:

Da _i ≦Tb _j+1, and, to respect deadline Eai,Db_j+Da_i≦Da_i+Ta_i, hence Db_j≦Ta_i

The conditions resulting from the situation of FIG. 4B are thus:

Da _i ≦Tb _j+1

AND

Db _j ≦Ta _i  [B.1]

From FIG. 5B, following the same reasoning as in FIG. 4B:

Da _i ≦Tb _j+1 +Tb _j+2

AND

Db _j +qlTb _j +Db _j+1 ≦Ta _i  [B.2]

As for condition A.1, the condition B.1 is verifiable in a situation where the frame Fa_i overlaps at least two consecutive frames Fb. The condition B.2 is verifiable in a situation where the frame Fa_i overlaps at least three consecutive frames Fb. In a similar manner, conditions B.3 and following may be found that are verifiable in situations where the frame Fa_i overlaps at least four or more consecutive frames Fb, but the likelihood of satisfying such conditions decreases rapidly. The conditions B.1 and B.2 (and, if applicable, A.3 and following) are alternative, that is to say, it is sufficient that any one of these conditions is satisfied for jumping to the validation of the other assumptions.

FIG. 6 illustrates exemplary frames that can be used in the development of compatibility criteria based on the assumption (C). In the context of assumption (C) it is assumed that operation Db_j has not been executed yet. If the deadline of the frame Fa_i arrives after the deadline of the frame Fb_j, the same situations as those of FIGS. 5A and 5B are encountered, where the operation Db_j just starts. Thus, under the assumption (C), only the case where the deadline of the frame Fa_i occurs before the deadline of frame Fb_j is analyzed.

Both operations Da_i and Db_j are ready for execution at the beginning of the frame Fa_i. It may be assumed that the operation Db_j was not started at the beginning of its frame, because the scheduling conditions 3) have not been met, for example, as shown, when the operation Da_i−1 ends at its deadline at the beginning of frame Fa_i. The deadline Eai being the shortest, the operation Da_i is launched first. The operation Db_j is launched at the end of the operation Da_i. In the worst-case within the limits of assumption (C), the deadlines of frames Fa_i and Fb_j coincide, which implies:

Db _j ≦Ta _i  [C.1]

This same inequality is obtained assuming operation Db_j starts at the beginning of the frame Fa_i, although the provided scheduling process would not proceed this way. For the operation Da_i to end before its deadline, the inequality Db_j+Da_i≦Da_i+Ta_i is verified, i.e. the same inequality as above.

Each frame Fa_i is thus compared to each frame Fb_j, ensuring that the deadlines of frames Fa_i are met, without however verifying that the deadlines of the frames Fb_j are met. This forms a first verification pass. A second pass to complete the verification is then to repeat the above operations by comparing each frame Fb_j to each frame Fa_i, thereby ensuring that the deadlines of frames Fb_j are met.

The verification of the compatibility of two repetitive sequences RSFa and RSFb having Na and Nb frames respectively, may be summarized by the following pseudo code. In this pseudo-code, alternative conditions are condensed in expressions min(x, y) at the left side of inequalities, and max(x, y) at the right side of inequalities and necessary conditions are condensed in expressions min(x, y) on the right side of inequalities, and max(x, y) on the left side of inequalities. Conditions A.3 and B.3 have been provided, as mentioned above, to reflect the event that frame Fa_i overlaps four frames Fb_j to Fb_j+3. Operations on the indices i and j are performed modulo the maximum value of the index (Na for i, and Nb for j), since the frame sequences are repeatable indefinitely.

For each i between 0 and Na−1 do:
For each j between 0 and Nb−1 do:
[A.1]
if Dai ≦ Tbj+1 then continue at [B.1]
[A.2]
ql := (Dai − Tbj+1)/Tbj
if ql > 1 or Tbj = 0 then ql := 1
if Dai ≦ Tbj+1 + Tbj+2
and qlTbj + Dbj+1 ≦ Tai
then continue at [B.1]
[A.3]
if Dai ≦ Tbj+2 + Tbj+3
and qlTbj + Dbj+1 + Dbj+2 + min[ Tbj+1, Dai ] ≦ Tai
then continue at [B.1]
exit the loop with failure code
[B.1]
if Dai ≦ Tbj+1
and Dbj ≦ Tai
then continue at [D]
[B.1.1]
ql := (Dai − Tbj+1)/Tbj
if ql > 1 or Tbj = 0 then ql := 1 and continue at [B.2]
Tl := Tai−1 − (1−ql)Tbj
if Dbj−1 ≦ Tl
then continue at [C.1]
[B.2]
if Dai ≦ Tbj+1 + Tbj+2
and Dbj + qlTbj + Dbj+1 ≦ Tai
then continue at [C.1]
[B.3]
if Dai ≦ Tbj+2 + Tbj+3
and Dbj + qlTbj + Dbj+1 + min[ Tbj+1, Dai ] + Dbj+2 ≦ Tai
then continue at [C.1]
exit the loop with failure code
[C.1]
if Dbj ≦ Tai
then continue at [D]
exit the loop with failure code
[D] next j
next i

To complete the verification, as previously indicated, this loop may be repeated switching the frames Fa and Fb.

In this loop, according to an alternative with respect to what has been described above, before proceeding with the evaluation of the condition B.2, a condition B.1.1 is evaluated.

FIG. 7 shows an exemplary succession of frames illustrating the establishment of the condition B.1.1. The frames Fa_i and Fb_j are in the same conditions as in FIG. 4A, but instead of starting the operation Da_i at the beginning of the frame Fb_j+1, it is assumed that it may start immediately after operation Db_j, and the general conditions that allow this with q=ql are sought in the previous frames Fa_i−1 and Fb_j−1. Condition B.1.1, as condition B.1 is based on the assumption that the frame Fa_i overlaps two frames (Fb_j and Fb_j+1), corresponding to a situation more likely to occur than one where the frame Fa_i overlaps three frames or more.

As previously indicated, the frames may include a reserve time R which may be taken by the system to perform non-real-time tasks. Such a reserve time may be used to perform real-time operations. It adds to the time margins T of one sequence to promote the execution of operations D of the other sequence, but it may harden the conditions for meeting deadlines.

Non-zero reserve times R may in particular relax the condition C.1.

FIG. 8 shows an exemplary succession of frames for establishing a relaxed constraint when the condition C.1 is not satisfied, that is to say if Db_j>Ta_i. Under the assumption C it is assumed that the operation Db_j has not yet started at the beginning of the frame Fa_i. The operation Da_i, having the shortest deadline, is started right at the beginning of the frame Fa_i. The frame Fb_j starts before the frame Fa_i, it is assumed during the interval Ra_i−1 of frame Fa_i−1. The interval between the beginning of the frame Fb_j and the beginning of the frame Fa_i is denoted kRa_i−1, with 0≦k≦1.

If the operation Db_j was not scheduled at the beginning of its frame Fb_j, it means that:

Db _j >kRa _i−1 +Ta _i.

The limit value of k is denoted kl, such that Db_j=klRa_i−1+Ta_i, hence kl=(Db_j−Ta_i)/Ra_i−1. If k>kl, it is certain that the operation Db_j can be scheduled before the beginning of the frame Fa_i—this case should not be considered because it does not fit the assumption C. Only the values of k between 0 and kl are considered.

For operation Da_i to be executable, it should meet the condition Da_i≦Tb_j−kRa_i−1. The worst case is k=kl, whereby:

Da _i ≦Tb _j −klRa _i−1, i.e., substituting kl by its value:

Da _i ≦Tb _j−(Db_j−Ta_i), or

Db _j ≦Ta _i +Tb _j −Da _i

The coefficient kl is at most equal to 1, which is expressed by:

Db _j −Ta _i ≦Ra _i−1, or

Db _j ≦Ta _i +Ra _i−1

Finally, to not constrain the execution of the operation Da_i+1, the operation Db_j should not overlap the frame Fa_i+1. This is satisfied if:

Db _j ≦Ta _i +Ra _i.

This leads to the following condition, an alternative to condition C.1:

Db _j ≦Ta _i +Tb _j −Da _i

AND

Db _j ≦Ta _i +Ra _i−1

AND

Db _j ≦Ta _i +Ra _i  [C.2]

The following pseudo-code takes into account the reserve times R.

For each i between 0 and Na−1 do:
For each j between 0 and Nb−1 do:
[A.1]
if Dai ≦ Tbj+1 then continue [B.1]
[A.2]
ql := (Dai − Tbj+1)/(Tbj + Rbj)
if ql > 1 ou Tbj + Rbj = 0 alors ql := 1
if Dai ≦ Tbj+1 + Rbj+1 + Tbj+2
and ql(Tbj + Rbj) + Dbj+1 ≦ Tai
then continue at [B.1]
[A.3]
if Dai ≦ Tbj+2 + Rbj+2 + Tbj+3
and Dbj+1 + ql(Tbj + Rbj)
+ Dbj+2 + min[ Tbj+1 + Rbj+1, Dai ] ≦ Tai
then continue at [B.1]
exit the loop with failure code
[B.1]
if Dai ≦ Rbj + Tbj+1
and Dbj ≦ Tai
then continue at [C.2]
ql := (Dai − Rbj − Tbj+1)/Tbj
if ql > 1 ou Tbj = 0 then ql := 1 and continue at [B.2]
[B.1.1]
Tl := Tai−1 − (1−ql)Tbj + Rai−1
if Dbj−1 ≦ Tl + min[ Tbj−1 + Dbj−1 + Rbj−1 − Tl − Dai−1, Rai−2 ]
then continue at [C.1]
[B.1.2]
if Dbj−1 + Rbj−1 ≦ Tl
and Dai−1 ≦ Tbj−2 + Rbj−2 + Tbj−1
and Tbj−2 + Dbj−2 + Rbj−2 + Tbj−1 + Dbj−1 + Rbj−1 ≦
Rai−2 + Tl + Dai−1
then continue at [C.1]
[B.2]
if Dai ≦ Tbj+1 + Rbj+1 + Tbj+2
and Dbj + qlTbj + Rbj + Dbj+1 ≦ Tai
then continue at [C.1]
[B.3]
if Dai ≦ Tbj+2 + Rbj+2 + Tbj+3
and Dbj + qlTbj + Rbj
+ Dbj+1 + min[ Tbj+1 + Rbj+1, Dai ]
+ Dbj+2 ≦ Tai
then continue at [C.1]
exit the loop with failure code
[C.1]
if Dbj ≦ Tai
then continue at [D]
[C.2]
if Dbj ≦ Tai + min[ Tbj−Dai, Rai, Rai−1 ]
then continue at [D]
exit the loop with failure code
[D] next j
next i

The condition B.1.2 that appears in this pseudo-code is determined using the same approach as for the condition B.1.1, extending searches to frames Fa_i−2 and Fb_j−2.

Until now, it was assumed that the time bases for clocking the two tasks were isochronous, that is to say that the durations of the intervals of one of the sequences were not variable relative to those of the other sequence. In practice, it may be necessary to compose two tasks using uncorrelated time bases, like tasks indexed to the positions of elements that move at variable speeds.

To handle this case, a variability coefficient z≧1 is introduced for each repetitive sequence, za and zb respectively for the sequences RSFa and RSFb. The coefficients za and zb are fixed and represent the maximum slowdown of the intervals T and R in the two sequences. Assuming that the notations used in the previous relationships represent the minimum values of the intervals, each of the values T and R may thus assume two values, T or zT and R or zR. The execution needs D remain constant because they only depend on the system clock.

In the previous relationships, where an interval is used to find a slot to execute an operation, its underestimated value T or R is maintained. When the interval is used to check compliance with a deadline, its overestimated value zT or zR is used. The following pseudo code is then obtained.

For each i between 0 and Na−1 do:
For each j between 0 and Nb−1 do:
[A.1]
if Dai ≦ Tbj+1 then continue at [B.1]
[A.2]
ql := (Dai − Tbj+1)/(Tbj + Rbj)
if ql > 1 or Tbj + Rbj = 0 then ql := 1
if Dai ≦ Tbj+1 + Rbj+1 + Tbj+2
and min[ ql(zb·(Tbj + Rbj) + (zb−1)Dbj), Dai ] + Dbj+1 ≦ Tai
then continue at [B.1]
[A.3]
if Dai ≦ Tbj+2 + Rbj+2 + Tbj+3
and Dbj+1 + min[ ql(zb·(Tbj + Rbj) + (zb−1)Dbj), Dai ]
+ Dbj+2 + min[ zb·(Tbj+1 + Rbj+1) + (zb−1)Dbj+1, Dai ] ≦ Tai
then continue at [B.1]
exit the loop with failure code
[B.1]
if Dai ≦ Rbj + Tbj+1
and Dbj ≦ Tai
then continue at [C.2]
[B.1.1]
ql := (Dai − Rbj − Tbj+1)/Tbj
if ql > 1 or Tbj = 0 then ql := 1 and continue at [B.2]
Tl := Tai−1 − (1−ql)Tbj + Rai−1
if Dbj−1 ≦ Tl + min[ Tbj−1 + Dbj−1 + Rbj−1 − Tl − Dai−1, Rai−2 ]
then continue at [C.1]
[B.1.2]
if Dbj−1 + zb·Rbj−1 ≦ Tl
and Dai−1 ≦ Tbj−2 + Rbj−2 + Tbj−1
and zb·(Tbj−2 + Dbj−2 + Rbj−2 + Tbj−1 + Dbj−1 + Rbj−1) ≦
Rai−2 + Tl +Dai−1
then continue at [C.1]
[B.2]
if Dai ≦ Tbj+1 + Rbj+1 + Tbj+2
and Dbj + min[ ql(zb·Tbj + (zb−1)Dbj) + zb·Rbj, Dai ]
+ Dbj+1 ≦ Tai
then continue at [C.1]
[B.3]
if Dai ≦ Tbj+2 + Rbj+2 + Tbj+3
and Dbj + min[ ql(zb·Tbj + (zb−1)Dbj) + zb·Rbj, Dai ]
+ Dbj+1 + min[ zb·(Tbj+1 + Rbj+1) + (zb−1)Dbj+1, Dai ]
+ Dbj+2 ≦ Tai
then continue at [C.1]
exit the loop with failure code
[C.1]
if Dbj ≦ Tai
then continue at [D]
[C.2]
if Dbj ≦ Tai + min[ Tbj−Dai, Rai, Rai−1 ]
then continue at [D]
exit the loop with failure code
[D] next j
next i

In case 3) of the scheduling process described above, the margin of time to consider is the minimum value, i.e. T and not zT.

The principles presented above have been described with two tasks to execute in timesharing, but they apply to any number of tasks. Indeed, the described method makes it possible in practice to compose a single sequencing plan for two compatible tasks. The sequencing plan thus obtained may be composed again, using the same techniques, with a sequencing plan for a third compatible task, and so on.

Claims

1. A method for executing two tasks in timesharing, comprising the steps of: decomposing offline each task in a repetitive sequence of consecutive frames in a time base associated with the task, wherein each frame is associated with an atomic operation having an execution need, and defines a start date from which the operation may start and a deadline by which the operation must end, whereby each frame defines a time margin in which the operation may start;verifying for each frame of a first of the repetitive sequences that the corresponding operation can be performed between any two successive operations of a group of frames of the second repetitive sequence, overlapping the frame of the first repetitive sequence, while respecting the start dates and deadlines of the operations; andif the verification is satisfied, allowing the execution of the two tasks and scheduling the operations of the two tasks according to the following cases, evaluated at each key date, wherein a key date occurs at each deadline, each start date, and each end of operation: if two operations can start, executing the operation having the shorter deadline; andif a single operation can start, executing it only if its execution need is less than the time remaining until the start date of the next frame of the other sequence, plus the margin of time associated with said next frame of the other sequence.

2. The method of claim 1, wherein each frame further comprises a reserve time interval that may be zero, located after the deadline of the frame or before the start date of the frame, wherein the reserve time intervals of any one of the repetitive sequences may be used to execute an operation of the other repetitive sequence.

3. The method of claim 2, wherein the verification step comprises the following steps: for each frame Fai of the first repetitive sequence, evaluating the following conditions for each frame Fbj of the second repetitive sequence: Dai≦Tbj+1  [A.1]Dai≦Rbj+Tbj+1  [B.1]and Dbj≦Tai, andDbj≦Tai,  [C.1]where D, T and R are the execution need, the time margin, and reserve time, indexed by a repetitive sequence identifier and the frame number in the sequence;repeating the condition evaluations with the frames Fai and Fbj exchanged;authorizing execution of the tasks if the three conditions [A.1], [B.1] and [C.1] are satisfied for any pair (i, j).

4. The method of claim 3, wherein the verification step comprises the following steps: if the condition [A.1] is false, evaluating the following condition: Dai≦Tbj+1+Rbj+1+Tbj+2  [A.2]and Dai+Dbj+1≦Tai+Tbj+1 if the condition [B.1] is false, evaluating the following condition: Dai≦Tbj+1+Rbj+1+Tbj+2  [B.2]and Dbj+Dai+Dbj+1≦Tai+Tbj+1 if the condition [C.1] is false, evaluating the following condition: Dbj≦Tai+min[Tbj−Dai,Rai,Rai−1]  [C.2]authorizing execution of the tasks if the conditions [A.1] or [A.2] and [B.1] or [B.2] and [C.1] or [C.2] are met for each pair (i, j).

5. The method of claim 3, wherein the time bases determining the durations of the frames are variable so that the time margins T and the reserve times R vary by a factor 1 to za for the frames Fa and a factor 1 to zb for the frames Fb, the verification step further comprising the steps of: if the condition [A.1] is false, evaluating the following condition: Dai≦Tbj+1+Rbj+1+Tbj+2  [A.2]and min[ql(zb·(Tbj+Rbj)+(zb−1)Dbj),Dai]+Dbj+1≦Tai where ql=(Dai−Tbj+1)/(Tbj+Rbj);if the condition [B.1] is false, evaluating the following condition: Dai≦Tbj+1+Rbj+1+Tbj+2  [B.2]and Dbj+min[ql(zb·Tbj+(zb−1)Dbj)+zb·Rbj,Dai]+Dbj+1≦Tai where ql=(Dai−Rbj−Tbj+1)/Tbj;if the condition [C.1] is false, evaluating the following condition: Dbj≦Tai+min[Tbj−Dai,Rai,Rai−1]  [C.2]authorizing execution of the tasks if the conditions [A.1] or [A.2] and [B.1] or [B.2] and [C.1] or [C.2] are met for each pair (i, j).

6. The method of claim 5, wherein the verification step comprises the following steps: if the condition [B.1] is false, evaluating the following condition before the condition [B.2]: Dai≦Tbj+Rbj+Tbj+1  [B.1.1]and Dbj−1≦Tl+min[Tbj−1+Dbj−1+Rbj−1−Tl−Dai−1,Rai−2]where ql=(Dai−Rbj−Tbj+1)/Tbj and Tl=(1−ql)Tbj+Rai−1 authorizing execution of the tasks if the conditions [A.1] or [A.2] and [B.1] or [B.1.1] and [C.1] or [C.2] are met for each pair (i, j).