CN109582448B - Criticality and timeliness oriented edge calculation task scheduling method - Google Patents
Criticality and timeliness oriented edge calculation task scheduling method Download PDFInfo
- Publication number
- CN109582448B CN109582448B CN201811208098.4A CN201811208098A CN109582448B CN 109582448 B CN109582448 B CN 109582448B CN 201811208098 A CN201811208098 A CN 201811208098A CN 109582448 B CN109582448 B CN 109582448B
- Authority
- CN
- China
- Prior art keywords
- task
- criticality
- timeliness
- consumption
- tasks
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F9/00—Arrangements for program control, e.g. control units
- G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
- G06F9/46—Multiprogramming arrangements
- G06F9/48—Program initiating; Program switching, e.g. by interrupt
- G06F9/4806—Task transfer initiation or dispatching
- G06F9/4843—Task transfer initiation or dispatching by program, e.g. task dispatcher, supervisor, operating system
- G06F9/4881—Scheduling strategies for dispatcher, e.g. round robin, multi-level priority queues
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F9/00—Arrangements for program control, e.g. control units
- G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
- G06F9/46—Multiprogramming arrangements
- G06F9/50—Allocation of resources, e.g. of the central processing unit [CPU]
- G06F9/5005—Allocation of resources, e.g. of the central processing unit [CPU] to service a request
- G06F9/5027—Allocation of resources, e.g. of the central processing unit [CPU] to service a request the resource being a machine, e.g. CPUs, Servers, Terminals
Landscapes
- Engineering & Computer Science (AREA)
- Software Systems (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mobile Radio Communication Systems (AREA)
- Data Exchanges In Wide-Area Networks (AREA)
Abstract
The invention discloses a criticality and timeliness-oriented edge calculation task scheduling method, which is characterized in that a network flow optimization method is utilized, criticality and effectiveness of different tasks in edge information processing are comprehensively considered, the criticality and timeliness-oriented edge calculation task scheduling method is provided, task processing resource consumption is calculated through criticality of information processing tasks input by users and time limit of task execution, a consumption graph model of the criticality and effectiveness of the tasks is established, and the network flow optimization method is adopted to schedule and allocate resources to the different tasks. The method can not only carry out fair resource allocation on the tasks, but also ensure that the tasks with high criticality and high effectiveness can be allocated with more resources to be calculated on the premise of fair resource allocation, and can be completed more quickly. The method is mainly suitable for task scheduling of the edge.
Description
Technical Field
The invention belongs to the technical field of task scheduling, and particularly relates to a criticality and timeliness oriented edge calculation task scheduling method.
Background
In an edge system, a large number of different types of services are often operated on different computing nodes, the different types of services process different types of tasks, the demands for computing resources are different, and if each task cannot be scheduled to a proper node, the different types of tasks can have relatively reasonable resources for computing, so that resource waste is caused, that is, the idle of the resources can still occur under the condition of a large amount of task backlogs, and the task response time is too long.
At present, with the advent of various distributed task scheduling technologies, designers of task scheduling systems need to solve the following problems in order to improve task scheduling efficiency and reduce system response time: 1) the scheduler cannot distinguish and treat long-term running service jobs and batch analysis jobs; 2) different types of scheduling strategy functions are required to be added when the scheduler processes different types of tasks, and the complexity of service logic and the complexity of configuration are increased, so that errors are easy to occur in deployment, operation and maintenance; 3) the scheduler may experience queue delay waiting and the task cannot roll back. In order to solve the above problems, researchers have proposed a network flow-based scheduling method.
However, the current scheduling algorithm based on network flow is mostly suitable for a computing service environment with sufficient cloud center resources and an application scenario without requirements on service execution time limit, and is not suitable for an edge system, the edge system is in a high-confrontation environment with variable battlefield environments, and important computing tasks are required to be completed by using limited resources within a limited time. If the traditional scheduling method is used in an edge system, the result is that all tasks are viewed equally, the tasks with high criticality and high timeliness cannot be allocated with enough resources, queue waiting is caused, and the tasks with low criticality and low timeliness occupy a large amount of limited resources. There is no guarantee that important edge tasks will be completed in a limited time. The specific problems of existing network flow based scheduling algorithms are as follows:
1) the consumption value calculation method of the method does not consider the task criticality and timeliness;
2) the resource allocation method of the method does not consider task criticality and timeliness;
3) the consumption function of the method does not take into account task criticality and timeliness.
Disclosure of Invention
The invention mainly aims to provide an edge task scheduling method based on task criticality and timeliness for an edge system in a high-confrontation and variable environment, so that an important edge computing task needing to be responded within a limited time can obtain enough computing resources, and the completion of the edge computing task is guaranteed preferentially. The invention specifically comprises the following steps:
step1, calculating a task criticality timeliness value of a task;
step2, calculating a task consumption value according to the task criticality timeliness value;
and 3, modifying the flow network diagram according to the data in operation, the task consumption value and the scheduling strategy.
And 4, reasonably distributing the resources acquired by the tasks according to the criticality and the timeliness.
The step1 comprises the following steps:
step 1-1, setting that the task has n different criticalities, recording the criticality of the ith task as Xi, defining the criticality of the task by a user, taking the value of i as 1-n, and calculating the relative criticality pi of the ith task according to the following formula:
step 1-2, determining the timeliness value Di of the ith task according to the required completion time Di of the ith task:
step 1-3, calculating a criticality timeliness value wi of the ith task according to the following formula:
wi＝pi*di；
step 1-4, carrying out minimum and maximum specification on the key timeliness values of all tasks through the following formula:
wherein maxa represents the maximum value of the criticality timeliness values of all tasks, mina represents the minimum value of the criticality timeliness values of all tasks, newmaxa、newminaRespectively representing the maximum and minimum values to be mapped to normalized, i.e. mapping wi values to newmina,newmaxa]In the interval, if newmina,newmaxa0 and 1, respectively, i.e. mapped in the interval 0 to 1.
The step2 comprises the following steps:
the consumption value wcosti for the ith task is defined as:
W1-Wn represents the weight of each dimension (CPU utilization dimension, memory utilization dimension, etc., and dimension information can be increased according to different scheduling policies), Wn represents the weight of the nth dimension, and the dimensions include CPU occupancy, memory occupancy, etc., CPU occupancy and memory occupancy are respectively represented by CPU and memory in formula (5), and dn represents the occupancy of other resources in the computer, and a user can define the occupancy according to the needs of the user, for example, if the required network bandwidth occupancy is minimum, dn can be the acquired task network bandwidth occupancy; the value range of each dimension is normalized to [0, c ], c is a constant value, the task related consumption is assigned to cost (u, v) when the graph is constructed, and the task related consumption cost (u, v) represents the consumption of the task which is executed after the task is dispatched to v by a computing node u and comprises the occupied cpu, the memory, the network bandwidth and the like.
The step 3 comprises the following steps:
and 3-1, representing the task scheduling flow network by using a directed graph G (V, E, U, C), wherein V is a node set, nodes represent resource supply and demand entities and comprise task demand parties and resource supply parties, and the nodes are used for representing resources used in scheduling tasks, aggregation nodes, racks, machines and the like. E is a set of edges, which represents that tasks are scheduled to corresponding machines for calculation; u represents the capacity of the edge, i.e., the maximum amount of resources that the machine can supply to accomplish a particular task; c represents the resource consumption needed by running the task after the task is scheduled to a specific machine;
step 3-2, searching all directed chains from a starting point sv to a terminal point tv in the directed graph G, sequentially marking the maximum capacity of each chain, and calculating corresponding minimum consumption by using a minimum consumption function facing to criticality and timeliness, if the maximum capacity is 0, the consumption is infinite, and marking the edge with the maximum capacity of 0 as | |, turning to step 3-3; otherwise, when all unit costs are 0, at this time, there is no longer a directed chain from sv to tv, and no further augmentation can be performed, at this time, the minimum cost is the maximum flow, and the scheduling is finished;
and 3-3, initializing the directed graph G to enable the initial feasible flow to be 0, selecting the directed chain with the minimum unit consumption, amplifying the directed chain with the maximum capacity, and ending scheduling when the directed chain from s v to t v does not exist any more, wherein the directed chain is the minimum consumption maximum flow.
In step 3-2, the minimum consumption function for criticality and timeliness is as follows:
min∑(w,v)∈Ewcost(w,v)f(w,v)，
wherein wcost (w, v) represents consumption of tasks in the task scheduling flow network from the computing node w to the computing node v for scheduling, and f (w, v) represents the number of tasks from w to v.
Step 4 comprises the following steps:
and setting the task set as T and the physical resource set as M, and mapping the task scheduling problem into a set of edges from the set T to the set M so as to minimize the consumption of the edges.
the scheme initially allocates ASj to each task from small to large according to resource requirements. If the allocation resources of the tasks exceed the required resources after the allocation is finished, the redundant resources are continuously allocated according to the formula.
In step 4, in order to determine the edge from T to M, the task scheduling constraint is described as the following triplet:
scheduling constraint ═ T, R, B >
Where T denotes the task, R denotes the resource that denotes the task's constraints, and B denotes the benefit that the task gets when scheduled on the resource. The invention adopts benefit function to describe scheduling constraint, namely:
max∑B (7)
the present invention translates the scheduling constraints into a problem of deciding whether to establish edges between different nodes of the graph. The present invention establishes an edge between T and R, T → R, each edge being assigned a cost value. cost is a relative value to benefit B. The minimum cost function corresponding to the maximum gain function is as follows
min∑cost (8)
The technical solution for achieving the object of the present invention comprises the following contents:
(1) the task criticality and timeliness calculation method comprises a task relative criticality calculation method, a timeliness calculation method and a criticality timeliness specification method. The task relative criticality calculation method calculates the ratio of each criticality according to the artificially set task criticality to calculate the relative criticality. Timeliness is calculated according to the service response time specification of the task, so that the less the time specification, the higher the timeliness. The influence of the criticality and the timeliness is comprehensively considered in a combined mode.
(2) The fair resource share distribution method based on the task criticality and the timeliness reasonably distributes resources acquired by the tasks according to the criticality and the timeliness; when confronted with the allocation of rare resources to a group of tasks, the tasks have equivalent rights to acquire resources, but the more critical and more time-efficient tasks should acquire more resources to ensure the efficient operation of the critical tasks. In this case, the task with relatively low criticality actually requires less resources than other users. The inventive solution solves how resources are allocated in this case. The invention provides a maximum and minimum fair resource sharing algorithm based on criticality and timeliness based on a resource sharing technology which is widely used in practical application and a maximum and minimum fair resource sharing algorithm. The algorithm allocates the unused resources to the tasks needing large resources according to the principle of priority of criticality and timeliness on the basis of fair sharing of the minimum requirements which can be met and are allocated to each task.
(3) The task criticality and timeliness-based consumption calculation method calculates the comprehensive criticality and effectiveness of each task through a task criticality and timeliness measurement method, and tasks with high criticality and high timeliness can obtain higher consumption in a task call flow diagram;
(4) the method for optimizing the minimum consumption maximum flow based on the task criticality and the timeliness is a core optimization process of the scheduling flow of the scheduling algorithm, and the scheduling flow is continuously optimized according to a consumption function based on the task criticality and the timeliness, so that tasks with high criticality and timeliness can be distributed to relatively large consumption. The optimization method finds a path from a source point to a sink point every time, increases the flow, and the path meets the requirement of minimizing the cost of the increased flow until a path from the source point to the sink point cannot be found, and the algorithm is finished, at this moment, the flow can reach the maximum flow of the network, and meanwhile, the key and timeliness consumption functions of the task allocate more flows for the key and highly-timeliness task; the final total consumption is minimal since each time there is a minimum cost increase, i.e. the current minimum cost is the minimum of all consumption when the current flow is reached.
Compared with the prior art, the invention has the following advantages:
(1) aiming at the problems that the current scheduling algorithm based on network flow is mostly suitable for a computing service environment with sufficient cloud center resources and an application scene without requirements on service execution time limit, is not suitable for an environment with high countermeasure and variable battlefield adaptation of a marginal system, and needs an important computing task to be completed by using limited resources in limited time, the patent uses a measurement method and a resource allocation method of task importance and timeliness, so that the task with high importance and timeliness obtains more resources, and waiting is reduced.
(2) Tasks with high importance and high timeliness are preferentially guaranteed in the scheduling process, so that the tasks can obtain more resources under the condition of fair resource distribution, and the tasks can be completed more quickly.
Drawings
The foregoing and other advantages of the invention will become more apparent from the following detailed description of the invention when taken in conjunction with the accompanying drawings.
FIG. 1 is a network flow based task scheduling for criticality and timeliness.
Fig. 2 is a flow network modeling diagram.
FIG. 3 is a flow network graph modeling abstraction.
Fig. 4 is a fair resource share allocation.
FIG. 5 is an example of task consumption values.
Detailed Description
The invention is further explained below with reference to the drawings and the embodiments.
The invention converts the task scheduling problem into a minimum consumption maximum flow optimization problem for a flow network based on task criticality and timeliness as shown in figure 1. And solving the minimum consumption maximum flow by combining the task criticality and the timeliness to obtain the optimal task scheduling position. I.e., scheduling the task to the most appropriate computing node for execution. So that the tasks with high criticality and high timeliness can be used by the tasks of different types most reasonably under the condition of guaranteeing the supply of the resources with the highest priority. The method comprises the following specific steps:
1. and calculating the criticality and timeliness of the task according to the task information.
The task information comprises specific content of the task, criticality of the task and timeliness of the task. The timeliness of a task means how long a task needs to be completed from the moment it is submitted, if this time is shorter, it means that the task is more urgent and more time-efficient, otherwise it is lower.
Let the task have n different criticalities, note that the criticality of each task is Xi, and higher values indicate higher criticalities. First, the relative criticality of each task is determined from the value of Xi, see equation (1):
and determining the timeliness value of each safety task according to the task required completion time Di:
determining the timeliness value of the mission criticality according to pi and di
wi＝pi*di (3)
In order to carry out calculation under the same standard, the invention carries out minimum-maximum specification after calculating the wi values of all tasks, and the minimum-maximum specification carries out linear transformation on the original values. Maximum maxa and minimum mina are assumed. Min-max normalization is calculated by:
wherein maxa represents the maximum value of the criticality timeliness values of all tasks, mina represents the minimum value of the criticality timeliness values of all tasks, newmaxa、newminaRespectively representing the maximum and minimum values to be mapped to normalized, i.e. mapping wi values to newmina,newmaxa]In the interval, if newmina,newmaxa0 and 1, respectively, i.e. mapped in the interval 0 to 1.
2. And calculating a task consumption value according to the task criticality timeliness value.
The method calculates the task consumption value wcost according to the task criticality timeliness value and the runtime monitoring values such as the task runtime cpu, the memory occupancy rate and the like. The consumption value wcosti for the ith task is defined as:
w 1-wn represents the weight of each dimension (CPU utilization dimension, memory utilization dimension, etc., dimension information can be added according to different scheduling strategies), the value range of each dimension is normalized to [0, c ], and c is a constant value. The task related consumption value is given as cost (u, v) when the graph is constructed.
3. The flow network graph is modified according to the runtime data, the calculated consumption values and the scheduling policy.
The task scheduling flow network is represented by a directed graph G (V, E, U, C), wherein V is a node set and represents resource supply and demand entities, including task demanders and resource supply, and nodes are used for representing resources used in scheduling tasks, aggregation nodes, racks, machines and the like. E is a set of edges representing the scheduling of tasks to the respective machines for computation. U represents the capacity of the edge, i.e. the amount of resources that the machine can supply at most in order to complete a particular task. C represents the resource consumption needed by the task to run after the task is scheduled to a specific machine.
The modeling result is shown in fig. 1, the left side is a task set T, the right side is a physical resource set M, M1 to M4, U1 and U2 represent tasks. The edge T0,2 → M1 indicates that T0,2 is scheduled to be executed on the M1 machine. T2,1 → U0 indicates that resource T2,1 is not scheduled for execution on the machine for the time being. The capacity of each edge indicates whether the task can be scheduled to a corresponding physical machine (capacity), and the consumption indicates that after the task is scheduled to the machine, more resources are required to be consumed to execute the task. The abstract notion of modeling of the present invention is shown in the following table and in fig. 2 and 3.
TABLE 1 modeling Abstract interpretation
The present invention maps the task scheduling problem into finding the set of edges from set T to set M, minimizing the consumption of edges. On the premise of ensuring the priority execution of the tasks with high importance and timeliness, the resource scheduling is more flexible, meanwhile, the task scheduling efficiency is improved, high task scheduling quality is obtained, the machine utilization rate is high, and the scheduling delay is low.
4. Determining waiting and scheduled tasks and deciding scheduling position
And determining the optimal solution of scheduling in the flow network by using a minimum consumption and maximum flow constraint solver based on criticality and timeliness, namely determining the optimal scheduling position and determining the tasks to be waited and scheduled in the flow network, and generating a new scheduling strategy for scheduling.
In order to find a set of edges from set T to set M, the present invention proposes a fair resource share allocation method based on task criticality and timeliness, which is used to first determine the edges from T to U, i.e., which tasks wait.
The fairness in the invention refers to the fact that the job can give consideration to the key degree and timeliness of the task on the basis of fairly sharing the resource share. And for the task j, the number of the tasks contained in the task j is NT, the fair share of the task j is calculated to be AS through a fair algorithm, and if the resource share distributed to the task j by the scheduling algorithm is AS, the scheduling algorithm meets the fairness. Fairness in the present invention is achieved by the graph edge capacity construction. As shown in fig. 4.
When confronted with the allocation of rare resources to a group of tasks, the tasks have equivalent rights to acquire resources, but the more critical and more time-efficient tasks should acquire more resources to ensure the efficient operation of the critical tasks. In this case, the task with relatively low criticality actually requires less resources than other users. The inventive solution solves how resources are allocated in this case. The invention provides a maximum and minimum fair resource sharing algorithm based on criticality and timeliness based on a resource sharing technology which is widely used in practical application and a maximum and minimum fair resource sharing algorithm. The algorithm allocates the unused resources to the tasks needing large resources according to the principle of priority of criticality and timeliness on the basis of fair sharing of the minimum requirements which can be met and are allocated to each task.
Consider a set of task services 1, …, n with resource requirements rs1, rs2, …, rsn, respectively. Assume rs1< ═ rs2< ═ … < ═ rsn, with the server having capability c. The resource requirement and the mission critical timeliness are comprehensively considered, and the resource ASj calculation method of the mission j provided by the invention comprises the following steps:
the scheme initially allocates ASj to each task from small to large according to resource requirements. If some tasks distribute resources more than needed resources after the distribution is finished, the redundant resources are continuously distributed according to the formula. After the process is finished, each task does not have more requirements than the task needs, and if the requirements are not met, the obtained resources are not less than the maximum resources obtained by other users. Tasks whose demand is not met will be in a waiting queue. The scheme maximizes the minimum allocated resources for users whose resources are not satisfied. Examples are as follows:
there is a user group G, which has 4 users, resource requirements of 2, 4, 4, 10, and weights of 4, 2.5, 1, 0.5, respectively, and total amount of resources of 16.
(1) The weights are first normalized, and with the minimum weight set to 1, the weights become 8, 5, 2,1, and the sum is 16. The total resource is divided into 16 equal parts, and four users respectively obtain 8, 5, 2 and 1.
(2) User 1 obtains 6 more resources, user 2 obtains 1 more resources, and users 3 and 4 do not meet the resources, so that the obtained 7 more resources are distributed to users 3 and 4 according to the weight, and users 3 and 4 respectively obtain 7 × 2/3 and 7 × 1/3 resources;
(3) so far, user 3 obtains 6.666 shares of resources, user 4 obtains 3.333, and the resources excess by the user are reallocated to user 4, thereby completing the allocation.
To determine the T to M edge, the present invention describes the task scheduling constraint as the following triplet:
scheduling constraint ═ T, R, B >
Where T denotes the task, R denotes the resource that denotes the task's constraints, and B denotes the benefit that the task gets when scheduled on the resource. The invention adopts benefit function to describe scheduling constraint, namely:
max∑B (7)
the present invention translates the scheduling constraints into a problem of deciding whether to establish edges between different nodes of the graph. The present invention establishes an edge between T and R, T → R, each edge being assigned a cost value. cost is a relative value to benefit B. The minimum cost function corresponding to the maximum gain function is as follows
min∑cost (8)
A simple example is now given to illustrate the impact of the consumption function on the schedule. As task T1 in fig. 5 is an image processing task that needs to be scheduled to a server with an image processor for execution, the two existing machines M1 and M2 now need to choose to schedule the image processing task to M1 or M2. M1 is configured with an image processor and M2 is not configured with an image-less processor. T1 has scheduling constraint on M1, and the obtained benefit is 1. T1 has no scheduling constraint on M2, with a benefit of 0. By calculating the minimum cost function, i.e., the maximum benefit function, T1 would be scheduled to M1 execution.
The invention provides a minimum consumption function for criticality and timeliness as follows:
min∑(w,v)∈Ewcost(w,v)f(w,v) (9)
where wcost (w, v) represents the consumption of tasks scheduled from w to v execution in the network flow graph, and f (w, v) represents the number of tasks from w to v.
The minimum cost maximum flow solving algorithm has the core idea of maintaining maximum flow and continuously and iteratively searching minimum cost, and the solving process is as follows:
step1 finds all the directed chains from the start point sv to the end point tv in the graph
(1) Finding all directed chains from sv to tv;
(2) sequentially marking the maximum capacity of each chain and calculating corresponding minimum consumption by using a minimum consumption function facing the criticality and the timeliness;
(3) if the maximum capacity is 0, the consumption is infinite, and the side having the maximum capacity of 0 is marked as "|",
step2 is switched; otherwise, when all unit costs are 0, at this time, there is no longer a directed chain from sv to tv, and there is no longer a possibility of
Performing augmentation, namely the maximum flow with the minimum cost at the moment, and ending;
step2 augments the initial feasible flow
(1) Initializing an original flow network diagram to enable an initial feasible flow to be 0;
(2) selecting the directed chain with the minimum unit consumption, and carrying out the maximum capacity augmentation on the directed chain;
(3) when the directional chain from s v to t v no longer exists, then no further augmentation can be performed, ending with the minimum consumption maximum flow.
The invention provides a criticality and timeliness oriented edge calculation task scheduling method, and a plurality of methods and ways for implementing the technical scheme are provided, the above description is only a preferred embodiment of the invention, it should be noted that, for a person skilled in the art, a plurality of improvements and embellishments can be made without departing from the principle of the invention, and these improvements and embellishments should also be regarded as the protection scope of the invention. All the components not specified in the present embodiment can be realized by the prior art.
Claims (4)
1. A criticality and timeliness oriented edge computing task scheduling method is characterized by comprising the following steps:
step1, calculating a task criticality timeliness value of a task;
step2, calculating a task consumption value according to the task criticality timeliness value;
step 3, modifying the flow network diagram according to the data at runtime, the task consumption value and the scheduling strategy;
step 4, reasonably distributing resources acquired by the tasks according to the criticality and the timeliness;
the step1 comprises the following steps:
step 1-1, setting that the task has n different criticalities, recording the criticality of the ith task as Xi, and calculating the relative criticality pi of the ith task according to the following formula, wherein i is 1-n:
step 1-2, determining the timeliness value Di of the ith task according to the required completion time Di of the ith task:
step 1-3, calculating a criticality timeliness value wi of the ith task according to the following formula:
wi＝pi*di；
step 1-4, carrying out minimum and maximum specification on the key timeliness values of all tasks through the following formula:
wherein maxa represents the maximum value of the criticality timeliness values of all tasks, mina represents the minimum value of the criticality timeliness values of all tasks, newmaxa、newminaRespectively representing the maximum and minimum values to be mapped to normalized, i.e. mapping wi values to newmina,newmaxa]In the interval, if newmina,newmaxa0 and 1, respectively, i.e. mapped in the interval 0 to 1;
the step2 comprises the following steps:
the consumption value wcosti for the ith task is defined as:
wherein W1-Wn represent the weight of each dimension, Wn represents the weight of the nth dimension, cpu occupancy and memory occupancy are respectively represented by cpu and memory in formula (5), and dn represents the occupancy of other resources in the computer; the value range of each dimension is normalized to [0, c ], c is a constant value, the task related consumption is assigned to cost (u, v) when the graph is constructed, and the task related consumption cost (u, v) represents the consumption of the task which is executed after the task is scheduled to v by a computing node u;
the step 3 comprises the following steps:
step 3-1, a task scheduling flow network is represented by a directed graph G (V, E, U, C), wherein V is a node set, and nodes represent resource supply and demand entities and comprise task demand parties and resource supply parties; e is a set of edges, which represents that tasks are scheduled to corresponding machines for calculation; u represents the capacity of the edge, i.e., the maximum amount of resources that the machine can supply to accomplish a particular task; c represents the resource consumption needed by running the task after the task is scheduled to a specific machine;
step 3-2, searching all directed chains from a starting point sv to a terminal point tv in the directed graph G, sequentially marking the maximum capacity of each chain, calculating corresponding minimum consumption by using a minimum consumption function facing to criticality and timeliness, if the maximum capacity is 0, the consumption is infinite, marking the side with the maximum capacity of 0 as | |, and turning to step 3-3; otherwise, when all unit costs are 0, at this time, there is no longer a directed chain from sv to tv, and no further augmentation can be performed, at this time, the minimum cost is the maximum flow, and the scheduling is finished;
and 3-3, initializing the directed graph G to enable the initial feasible flow to be 0, selecting the directed chain with the minimum unit consumption, amplifying the directed chain with the maximum capacity, and ending scheduling when the directed chain from s v to t v does not exist any more, wherein the directed chain is the minimum consumption maximum flow.
2. The method of claim 1, wherein in step 3-2, the criticality and timeliness oriented minimum cost function is as follows:
min∑(w,v)∈Ewcost(w,v)f(w,v)，
wherein wcost (w, v) represents consumption of tasks in the task scheduling flow network from the computing node w to the computing node v for scheduling, and f (w, v) represents the number of tasks from w to v.
3. The method of claim 2, wherein step 4 comprises:
setting a task set as T, a physical resource set as M, and mapping a task scheduling problem into a set of edges from the set T to the set M so as to minimize the consumption of the edges;
the task service sets 1 and 2 … are set, n has resource requirements rs1, rs2, … and rsn respectively, rs1< ═ rs2< ═ … < ═ rsn is set, the server has the capability rc, and the resource ASj calculation method of the task j is as follows:
and allocating the ASj to each task from small to large according to the resource requirement, and if the allocated resources of the tasks exceed the required resources after the allocation is finished, continuously allocating the redundant resources according to a formula (6).
4. The method of claim 3, wherein in step 4, to determine the T to M edge, the task scheduling constraint is described as the following triplet:
the scheduling constraint is < T, R, B >,
wherein T represents a task, R represents a resource representing a task constraint, B represents a benefit obtained when the task is scheduled to the resource, and a benefit function is adopted to describe the scheduling constraint, namely:
max∑B (7)
converting the scheduling constraint into a problem of determining whether to establish an edge between different nodes of the graph, establishing an edge between T and R → R, each edge being assigned a cost value, the cost being a value corresponding to the benefit B, and the minimum consumption function corresponding to the maximum benefit function being as follows:
min∑cost (8)。
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811208098.4A CN109582448B (en) | 2018-10-17 | 2018-10-17 | Criticality and timeliness oriented edge calculation task scheduling method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811208098.4A CN109582448B (en) | 2018-10-17 | 2018-10-17 | Criticality and timeliness oriented edge calculation task scheduling method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109582448A CN109582448A (en) | 2019-04-05 |
| CN109582448B true CN109582448B (en) | 2021-03-16 |
Family
ID=65920628
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811208098.4A Active CN109582448B (en) | 2018-10-17 | 2018-10-17 | Criticality and timeliness oriented edge calculation task scheduling method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109582448B (en) |
Families Citing this family (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110198339B (en) * | 2019-04-17 | 2020-08-21 | 浙江大学 | QoE (quality of experience) perception-based edge computing task scheduling method |
| CN110099384B (en) * | 2019-04-25 | 2022-07-29 | 南京邮电大学 | Multi-user multi-MEC task unloading resource scheduling method based on edge-end cooperation |
| CN110515716B (en) * | 2019-08-29 | 2021-11-30 | 中国科学院软件研究所 | Cloud optimization scheduling method and system supporting priority and inverse affinity |
| CN110532078A (en) * | 2019-08-29 | 2019-12-03 | 中国科学院软件研究所 | A kind of edge calculations method for optimizing scheduling and system |
| CN110647398A (en) * | 2019-09-18 | 2020-01-03 | 北京航空航天大学 | Intersection control task scheduling method facing edge calculation and based on task criticality and timeliness |
| CN110995855B (en) * | 2019-12-13 | 2022-06-21 | 深圳先进技术研究院 | Microservice cluster scheduling method, scheduling device and computer readable storage medium |
| CN112466115A (en) * | 2019-12-30 | 2021-03-09 | 北京航空航天大学 | Bus intersection priority passing control system and method based on edge calculation |
| CN111858051B (en) * | 2020-07-20 | 2023-09-05 | 国网四川省电力公司电力科学研究院 | Real-time dynamic scheduling method, system and medium suitable for edge computing environment |
| CN112291335B (en) * | 2020-10-27 | 2021-11-02 | 上海交通大学 | Optimized task scheduling method in mobile edge calculation |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103324525A (en) * | 2013-07-03 | 2013-09-25 | 东南大学 | Task scheduling method in cloud computing environment |
| CN103838621A (en) * | 2012-11-27 | 2014-06-04 | 中国电信股份有限公司 | Method and system for scheduling routine work and scheduling nodes |
| CN105677461A (en) * | 2015-12-30 | 2016-06-15 | 西安工业大学 | Mixed-criticality tasks scheduling method based on criticality |
| CN105868190A (en) * | 2015-01-19 | 2016-08-17 | 中国移动通信集团河北有限公司 | Method and system for optimizing task processing in ETL |
-
2018
- 2018-10-17 CN CN201811208098.4A patent/CN109582448B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103838621A (en) * | 2012-11-27 | 2014-06-04 | 中国电信股份有限公司 | Method and system for scheduling routine work and scheduling nodes |
| CN103324525A (en) * | 2013-07-03 | 2013-09-25 | 东南大学 | Task scheduling method in cloud computing environment |
| CN105868190A (en) * | 2015-01-19 | 2016-08-17 | 中国移动通信集团河北有限公司 | Method and system for optimizing task processing in ETL |
| CN105677461A (en) * | 2015-12-30 | 2016-06-15 | 西安工业大学 | Mixed-criticality tasks scheduling method based on criticality |
Non-Patent Citations (1)
| Title |
|---|
| 基于协同效益的任务与资源匹配优化方法;郭成昊 等;《指挥信息系统与技术》;20150630;第6卷(第3期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109582448A (en) | 2019-04-05 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109582448B (en) | Criticality and timeliness oriented edge calculation task scheduling method | |
| US11275609B2 (en) | Job distribution within a grid environment | |
| CN110825520B (en) | Cluster extremely-fast elastic telescoping method for realizing efficient resource utilization | |
| CN107688492B (en) | Resource control method and device and cluster resource management system | |
| CN110308967B (en) | Workflow cost-delay optimization task allocation method based on hybrid cloud | |
| CN109783225B (en) | Tenant priority management method and system of multi-tenant big data platform | |
| CN108270805B (en) | Resource allocation method and device for data processing | |
| CN105718316A (en) | Job scheduling method and apparatus | |
| CN103257896B (en) | A kind of Max-D job scheduling method under cloud environment | |
| US20120324466A1 (en) | Scheduling Execution Requests to Allow Partial Results | |
| Sharma et al. | An improved task allocation strategy in cloud using modified k-means clustering technique | |
| Thirumalaiselvan et al. | A strategic performance of virtual task scheduling in multi cloud environment | |
| CN111159859B (en) | Cloud container cluster deployment method and system | |
| Pandya et al. | Dynamic resource allocation techniques in cloud computing | |
| Singh et al. | Value and energy optimizing dynamic resource allocation in many-core HPC systems | |
| He et al. | Dynamic scheduling of parallel real-time jobs by modelling spare capabilities in heterogeneous clusters | |
| US20240004707A1 (en) | Methods and systems for energy-efficient scheduling of periodic tasks on a group of processing devices | |
| CN110308991A (en) | A kind of data center's energy conservation optimizing method and system based on Random Task | |
| Tahir et al. | UDRF: Multi-resource fairness for complex jobs with placement constraints | |
| Kanemitsu et al. | Prior node selection for scheduling workflows in a heterogeneous system | |
| Chatterjee et al. | A multi-objective deadline-constrained task scheduling algorithm with guaranteed performance in load balancing on heterogeneous networks | |
| CN112306642B (en) | Workflow scheduling method based on stable matching game theory | |
| Chouhan et al. | A MLFQ Scheduling technique using M/M/c queues for grid computing | |
| Komarasamy et al. | Deadline constrained adaptive multilevel scheduling system in cloud environment | |
| Rahman et al. | Group based resource management and pricing model in cloud computing |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |