CN115421901A - Priority perception task scheduling method and system for computational power network - Google Patents
Priority perception task scheduling method and system for computational power network Download PDFInfo
- Publication number
- CN115421901A CN115421901A CN202210929252.7A CN202210929252A CN115421901A CN 115421901 A CN115421901 A CN 115421901A CN 202210929252 A CN202210929252 A CN 202210929252A CN 115421901 A CN115421901 A CN 115421901A
- Authority
- CN
- China
- Prior art keywords
- task
- scheduling
- tasks
- priority
- computing
- 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.)
- Pending
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
- G06F9/50—Allocation of resources, e.g. of the central processing unit [CPU]
- G06F9/5083—Techniques for rebalancing the load in a distributed system
-
- 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
- G06F9/44—Arrangements for executing specific programs
- G06F9/455—Emulation; Interpretation; Software simulation, e.g. virtualisation or emulation of application or operating system execution engines
- G06F9/45533—Hypervisors; Virtual machine monitors
- G06F9/45558—Hypervisor-specific management and integration aspects
-
- 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
-
- 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
- G06F9/44—Arrangements for executing specific programs
- G06F9/455—Emulation; Interpretation; Software simulation, e.g. virtualisation or emulation of application or operating system execution engines
- G06F9/45533—Hypervisors; Virtual machine monitors
- G06F9/45558—Hypervisor-specific management and integration aspects
- G06F2009/4557—Distribution of virtual machine instances; Migration and load balancing
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management
Landscapes
- Engineering & Computer Science (AREA)
- Software Systems (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Data Exchanges In Wide-Area Networks (AREA)
Abstract
The invention provides a priority perception task scheduling method and a system facing a computational power network, wherein the method comprises the following steps: the method comprises the steps that a computing force controller periodically collects task arrival information of a computing force router and virtual machine occupation states of different computing force service nodes, wherein the task arrival information is information of tasks to be scheduled in a task buffer queue; the computing force controller analyzes the collected task arrival information and the virtual machine occupation state; evaluating the scheduling priority of each task according to a preset priority determination algorithm, and determining the schedulable task in the current period by taking the number of idle virtual machines as a limit; obtaining a scheduling decision through a predetermined comprehensive optimization target of task scheduling based on the determined schedulable task; and based on the obtained scheduling decision, the computation force router executes corresponding scheduling action, dispatches the schedulable tasks to the corresponding computation force service nodes for processing, and continues to queue the rest low-priority tasks for waiting for next scheduling.
Description
Technical Field
The invention relates to the technical field of computing power networks, in particular to a priority perception task scheduling method and system for a computing power network.
Background
With the advent of the big data age, mobile data generated from various distributed sources has presented a trend of explosive growth. Edge computing has been increasingly emphasized as a computing paradigm with great potential. Compared with cloud computing, edge computing deploys computing resources at the edge of a network closer to a data source, and constructs an open platform integrating core functions such as network, computing and storage, so that propagation delay caused by geographic distance in cloud computing can be eliminated, and computing services with low delay, low energy consumption and high security level can be provided for users. However, with the explosion of everything interconnection, the next generation of compute-intensive and delay-sensitive services places higher demands on the flexibility and processing power of the network. Further development of edge computing suffers from some inherent limitations due to its single point resource limited nature. When the task load among a plurality of edge nodes in an edge computing system is unevenly distributed, not only can serious network congestion be caused, but also the available computing resources of some edge nodes in an idle state can be wasted. Through resource sharing among different edge nodes, edge cooperation is considered as a promising method for solving single-point bottleneck and promoting edge network load balancing, and can fully utilize task processing capacity of an edge layer while exerting edge processing advantages.
Existing research related to edge collaboration focuses mainly on small-scale task scheduling and resource allocation between adjacent nodes within a certain domain. In the existing computation offloading or task scheduling scheme related to edge computation, the computation resources at each edge computation node may be regarded as a resource pool, and when performing a task scheduling decision, the decision optimization of computation offloading or task scheduling between neighboring nodes is performed usually with time or energy consumption as an optimization target, taking into consideration the condition of resource consumption required for processing the computation task of the user terminal, the state of the computation resources available at the edge computation node, the change of the topology of the edge computation network, and the like. And placing the task to the adjacent node cooperation process. However, with the ever-expanding network size and the continuing increase in network complexity, large-scale collaboration between multiple heterogeneous edge nodes across operators or domains becomes very challenging. Most of the existing edge computing task scheduling schemes concern hierarchical task unloading between different hierarchies, namely cloud edge terminals, and detailed research on horizontal task scheduling in an edge network is lacked. When the workload distribution among a plurality of edge nodes in the edge network is uneven, if only the hierarchical offloading within a certain proximity is considered, a large amount of edge idle resources are wasted, and the task processing capability of the edge network is not fully utilized. Meanwhile, most of the existing horizontal task scheduling schemes in the edge layer are limited to cooperative task processing between adjacent nodes in a single domain, and with the development trend that multi-dimensional resource super-fusion networking becomes a future network, the existing small-scale cooperative task scheduling schemes are difficult to meet the task processing requirements under the cross-domain complex networking environment, and the research on how to realize efficient resource sharing and cooperative task processing among multiple heterogeneous nodes in the edge layer is urgently needed. The advent of computational networks has provided a promising solution to the above challenges. In 2021, the international telecommunication union has formally released standard ITU-ty.2501 to describe architecture of computational power networks, which is a novel network for flexibly allocating resources such as computation, storage, and networks of each service node through a network control plane to realize efficient resource sharing. The computational network can connect the computing nodes scattered at the edge of the network, and has strong competitiveness in the aspect of realizing efficient edge cooperation of a large-scale edge computing system due to huge potential in the aspects of on-demand scheduling and flexible resource sharing. As a possible next generation network organization, the computing power network has received a great deal of attention from the academic community. Currently, the research on computational power networks is still in a starting stage, and most of the existing researches analyze the development trend of network and computation fusion and the key technical problems of related protocol design, energy consumption optimization and the like from an idealized perspective. However, from the perspective of a service provider, how to adaptively schedule tasks according to specific task conditions to realize multi-heterogeneous node cooperation by a computational network control plane is a key problem in practical applications, but has not been studied in detail, while ensuring system load balance and realizing efficient computing resource sharing. In addition, the existing task scheduling scheme mostly ignores the personalized characteristics of different tasks, and with the continuous development of computation-intensive and delay-sensitive intelligent applications, according to the characteristics of the different tasks, different scheduling priorities are divided to perform adaptive flexible scheduling to realize more comprehensive task processing capability of the network, which needs further consideration and research.
Disclosure of Invention
In view of this, the embodiment of the present invention provides a priority-aware task scheduling method and system for a computational power network, and combines an edge computing technology and a computational power network technology to design a novel integrated architecture for multi-node cooperation of the computational power network, so as to implement effective fusion of computation and the network, flexible sharing of computation resources, a higher task processing success rate, and full utilization of network resources.
One aspect of the present invention provides a priority-aware task scheduling method for a computational power network, including the following steps:
the computational force controller periodically collects task arrival information of the computational force router and virtual machine occupation states of different computational force service nodes, wherein the task arrival information is information of a task to be scheduled in a task buffer queue;
the computing force controller analyzes the collected task arrival information and the virtual machine occupation state; evaluating the scheduling priority of each task according to a preset priority determining algorithm, and determining the schedulable task of the current period by taking the number of idle virtual machines as a limit;
obtaining a scheduling decision through a comprehensive optimization target of scheduled task scheduling based on the determined schedulable task;
and based on the obtained scheduling decision, the computation force router executes corresponding scheduling action, dispatches the schedulable tasks to the corresponding computation force service nodes for processing, and continues to queue the rest low-priority tasks for waiting for next scheduling.
In some embodiments of the invention, there is a buffer queue on the computational router that is used to temporarily hold tasks that are temporarily not scheduled.
In some embodiments of the invention, the method further comprises:
based on the limited length of the buffer queue, when the buffer queue is full, the rest tasks which cannot enter the queue are directly dispatched to the cloud center.
In some embodiments of the present invention, the predetermined priority determination algorithm determines the scheduling priority of the task to be scheduled in the system by performing priority scoring on the task to be scheduled based on the combination of the delay requirements of each task, the total task load reaching the computation router, and the time of the task reaching the system; the priority score is expressed as:
wherein the content of the first and second substances,
indicating the arrival of the ith computation-force route at the t slotDevice r i And queuing the kth task to be scheduled, T, in its buffer queue 0 In order to schedule the length of the time slot,
denoted as a task to be scheduled
Numi (t) is expressed as the total task load to reach the ith computational router at time slot t, o t Denoted as system time at the start of the t slot, o τ Denoted as a task to be scheduled
Time of arrival in the system, beta 1 、β 2 、β 3 The weight factors of the three types of influence factors when evaluating the priority.
In some embodiments of the present invention, the number of idle virtual machines is used as a limit to determine that the total number of schedulable tasks in the current period is not greater than the number of idle virtual machines in the system.
In some embodiments of the present invention, the comprehensive optimization goal of task scheduling is the total task processing delay and the load balancing condition of the system; the scheduling decision is to select proper computational power service nodes for different schedulable tasks;
the total task processing time delay comprises the transmission time delay of the task and the execution time delay of the task, and for the task which is not scheduled in the scheduling time slot, the total task processing time delay also comprises the time delay of waiting for scheduling of the task;
for the k-th task to be scheduled which arrives at the ith computation force router at the t time slot and is queued in the buffer queue of the ith computation force router
The various delays are represented as follows:
the transmission delay is expressed as
Wherein the content of the first and second substances,
representing tasks
Actual transmission data size of R ij Entry computation force router r representing arrival of current task i With the final serving node e j The data transmission rate therebetween;
the execution delay is expressed as
Wherein, the first and the second end of the pipe are connected with each other,
representing tasks
The calculated amount of (1), F j Is a service node e j Total task processing capacity, V j Representing the number of virtual machines deployed by the service node;
the latency is expressed as
Wherein, T 0 The length of a scheduling time slot is m, and the m represents the scheduling turn of waiting of the current task; if the priority is higher, it is not necessary toWaiting for the next round of scheduling, and then m =0;
the total task processing delay of the current task is represented as:
when i = j, it means that the current task is processed at the computation force service node directly connected to the computation force router without being scheduled to other service nodes.
In some embodiments of the invention, the computing service nodes directly connected to the computing routers are ingress service nodes, and each computing router has a computing service node directly connected to it.
In some embodiments of the present invention, the method further comprises evaluating a load balancing condition of the network by using the variance of the virtual machine occupancy rates of the computing power service nodes after the scheduling is completed.
Another aspect of the present invention provides a priority-aware task scheduling system for a computing power network, comprising a processor and a memory, wherein the memory stores computer instructions, and the processor is configured to execute the computer instructions stored in the memory, and when the computer instructions are executed by the processor, the system implements the steps of the method as described above.
The invention relates to a priority perception task scheduling method and a system facing a computing power network, which can design a novel integrated architecture facing multi-node cooperation of the computing power network by combining an edge computing technology and a computing power network technology, further clarify the organization relation among functional entities under the architecture and the communication interaction flow of the computing system, consider the heterogeneity and resource limitation characteristics of computing power service nodes, limit the maximum processable task number of the system in real time, distinguish the scheduling priorities of different tasks according to the respective characteristics of the tasks for flexible scheduling, design an efficient task scheduling mechanism from the perspective of a service provider, ensure the load balance of the system while meeting the low delay requirement of a user, and realize the full utilization of network resources, the effective fusion of computing and the network, the flexible sharing of computing resources and higher task processing success rate.
Additional advantages, objects, and features of the invention will be set forth in part in the description which follows and in part will become apparent to those having ordinary skill in the art upon examination of the following or may be learned from practice of the invention. The objectives and other advantages of the invention will be realized and attained by the structure particularly pointed out in the written description and drawings.
It will be appreciated by those skilled in the art that the objects and advantages that can be achieved with the present invention are not limited to the specific details set forth above, and that these and other objects that can be achieved with the present invention will be more clearly understood from the detailed description that follows.
Drawings
The accompanying drawings, which are included to provide a further understanding of the invention and are incorporated in and constitute a part of this application, illustrate embodiment(s) of the invention and together with the description serve to explain the principles of the invention. In the drawings:
fig. 1 is a multi-node cooperative integration architecture and a communication flow diagram of a computational power network according to an embodiment of the present invention.
Fig. 2 is a schematic flowchart of a priority-aware task scheduling method for a computational power network according to an embodiment of the present invention.
Fig. 3 is a flowchart of communication interaction in a multi-node cooperative integrated architecture of a computational power network according to an embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is further described in detail below with reference to the following embodiments and the accompanying drawings. The exemplary embodiments and descriptions of the present invention are provided to explain the present invention, but not to limit the present invention.
It should be noted that, in order to avoid obscuring the present invention with unnecessary details, only the structures and/or processing steps closely related to the scheme according to the present invention are shown in the drawings, and other details not so relevant to the present invention are omitted.
It should be emphasized that the term "comprises/comprising" when used herein, is taken to specify the presence of stated features, elements, steps or components, but does not preclude the presence or addition of one or more other features, elements, steps or components.
It is also noted herein that the term "coupled," if not specifically stated, may refer herein to not only a direct connection, but also an indirect connection in which an intermediate is present.
Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings. In the drawings, the same reference numerals denote the same or similar components, or the same or similar steps.
The method aims to solve the problems that in the prior art, detailed research on horizontal task scheduling in an edge network is lacked, the task processing requirements under a cross-domain complex networking environment are difficult to meet, the personalized characteristics of different tasks are ignored, and the like. The invention provides a priority perception task scheduling method and system for an computational power network, and the method is combined with an edge computing technology and a computational power network technology to design a novel computational power network multi-node cooperation-oriented integrated architecture, so that the organization relation among functional entities under the architecture and the communication interaction flow of a computing system are further defined. Considering the isomerism of the computing service node and the resource limitation characteristic, the maximum number of tasks capable of being processed by the system in real time is taken as the limitation, the scheduling priorities of different tasks are distinguished according to the respective characteristics of the tasks for flexible scheduling, and from the perspective of a service provider, an efficient task scheduling mechanism is designed to meet the low delay requirement of a user and guarantee the load balance of the system at the same time, so that the full utilization of network resources, the effective fusion of computing and a network, the flexible sharing of computing resources and higher task processing success rate are realized.
In an embodiment of the present invention, a computing power network multi-node cooperation integrated architecture is designed based on an edge computing technology and a computing power network technology, as shown in fig. 1, the architecture is composed of a plurality of heterogeneous computing power service nodes, a centralized computing power controller, a plurality of computing power routers, a cloud computing center, and a plurality of common routers only responsible for data forwarding, and specific functions of each entity are described as follows:
and the computing power service node is responsible for task processing in the system. Different computing power service nodes establish a virtualized computing pool by deploying virtual mechanisms, so that the on-demand distribution of computing resources in the system can be realized; the parallel processing capacity of each computational service node is different according to the number of virtual machines deployed.
The computing force controller provides a global network control and resource arrangement function for the system, and can be communicated with different computing service nodes and computing routers regularly to collect system states and issue scheduling decisions; the computational controller has a global overview of all tasks and available resources in the computing system, enabling global management of system scheduling actions.
The computational power router is a network device which can sense the arrival of tasks in the system and schedule the tasks according to the decisions sent by the computational power controller. In addition, the system also has a buffer queue to temporarily store the temporarily unscheduled tasks, and the scheme assumes that the upper limit of the length of all queues is L. Each computation force router has a computation force service node directly connected with the computation force router, and particularly, for the task of reaching a certain computation force router, the computation force service node directly connected with the computation force router can be regarded as an entrance service node.
The cloud computing center can directly process two types of tasks; the first kind of tasks is that because the length of a task buffer queue is limited, when the buffer queue is full, some tasks which cannot be enqueued may exist; the second category of tasks is that with the continued development of intelligent applications there may be computationally intensive tasks that some edge service nodes are difficult to handle. The two types of tasks can be directly scheduled to the cloud center for processing.
Fig. 2 is a schematic flow chart of a priority-aware task scheduling method for a computational power network according to an embodiment of the present invention, as shown in fig. 2, the method includes the following steps:
step S110, the computing force controller periodically collects task arrival information of the computing force router and virtual machine occupation states of different computing force service nodes, wherein the task arrival information is task information to be scheduled in a task buffer queue.
In this step, the computing force controller periodically communicates with different computing force service nodes and computing force routers, and periodically collects task arrival information of the computing force routers and virtual machine occupation states of the different computing force service nodes. The computational router is used as a network device, receives tasks in the system and temporarily stores the tasks which are not scheduled temporarily by a buffer queue; the computing service node establishes a virtualized computing pool by deploying a virtual mechanism to take charge of task processing in the system.
In an embodiment of the invention, based on the limited length of the buffer queue, if a task buffer queue of a certain computation router is full, other tasks which cannot enter the queue are directly scheduled to the cloud center for processing.
Step S120, the computing force controller analyzes the collected task arrival information and the virtual machine occupation state; and evaluating the scheduling priority of each task according to a preset priority determination algorithm, and determining the schedulable task in the current period by taking the number of idle virtual machines as the limit.
In this step, the computing force controller analyzes according to the collected task arrival information and virtual machine occupation states, in order to improve task processing efficiency of the collaborative computing system and avoid occurrence of network congestion, in an embodiment of the present invention, a "sliding window" mechanism in a TCP protocol is used for reference, and it is constrained that the schedulable number of tasks in each scheduling time slot is consistent with the total number of idle virtual machines on heterogeneous computing power service nodes in the system, that is, there is a one-to-one correspondence between the scheduled tasks and the idle virtual machines, so that it can be ensured that all schedulable tasks can be directly processed without queuing again after reaching the service nodes. However, according to different system states, there are two possible scheduling cases, scheduling case one: if the number of the tasks to be scheduled which arrive at a certain time slot in the system is less than or equal to the number of the idle virtual machines in the system at the moment, all the arrived tasks can be scheduled; and B, scheduling condition two: when the number of the tasks to be scheduled is larger than the number of the idle virtual machines at the moment, a part of the tasks cannot be scheduled in the time slot. Therefore, the scheduling priorities of different tasks in the system need to be determined according to a priority algorithm, and high-priority tasks need to be screened out. The tasks with high priority can be scheduled in the time slot, and other tasks with lower priority need to be queued in the queue in sequence for scheduling. And after the high-priority tasks are screened out, the optimal scheduling decision is obtained according to the comprehensive optimization target of task scheduling and is issued to each computational power router.
In the embodiment of the invention, based on a computing power network multi-node cooperation integrated structure, a priority perception dynamic task scheduling scheme facing edge cooperation is designed by taking the real-time maximum processable task total number of a system as a constraint condition, and the scheme divides a task scheduling problem into two sub-problems of determining the scheduling priority of different tasks and selecting proper computing power service nodes for the different tasks.
For the sub-problem one: the scheduling priorities of the different tasks are determined. In order to distinguish the scheduling priorities of different tasks in the system, more flexible resource sharing can be realized, individual requirements of more users are met by allocation according to needs, the upper limit of the schedulable task number of the time slot can be determined by taking the total number of the real-time idle virtual machines of the whole system as a constraint, namely the total number of the finally scheduled tasks Mum in the t-time-slot edge system sche (t) is not more than the total number Num of idle virtual machines VMidle (t) that is
Mum sche (t)≤Num VMidle (t)；
The algorithm 1 is a scheduling priority determining algorithm for different tasks to be scheduled in the system provided by an embodiment of the present invention, and the task scheduling priority determining algorithm 1 comprises the following steps:
the requirement for time delay is greater than T 0 And the tasks with undetermined priority are sorted according to the priority scores;
before taking (Num) VMidle (t) -count) tasks and setting the scheduling priority of the tasks to 1;
the scheduling priority of the remaining tasks is set to 0.
Less than system scheduling slot length T for some delay requirements 0 And reaches the ith computation force router r in the t time slot i I ∈ {1,2, \8230;, N }, and queued in its task buffer queue for the kth task to be scheduled
When its scheduling priority
Then, the scheduling can be carried out in the time slot; its scheduling priority
Then, it needs to wait for scheduling. In one embodiment of the present invention, assuming that the number of tasks of this type does not exceed the total number of idle virtual machines in the system, the requirement for the remaining delay is greater than the scheduling slot length T 0 The determination of the scheduling priority needs to be balanced in multiple aspects according to the task characteristics to ensure the relative fairness of the scheduling decision. On one hand, considering the uneven distribution of the number of the tasks to be scheduled of each entry computing force router, the priority of task scheduling can be determined according to the number of the tasks reaching each computing force router, namely, the more the tasks to be scheduled are queued by a certain computing force router at the time slot, the more the high-priority tasks which can be scheduled on the router are; on the other hand, from the perspective of system global, the task that arrives first should have higher scheduling priority based on the difference of the arrival order of each task in the system. In summary, for tasks with delay requirements greater than the length of the scheduling slot
In particular, the delay requirements of the task need to be taken into account comprehensively
Arriving at computation-force router r at t time slot i Total task load Num of i (t) and the time o at which the task to be scheduled arrives at the computing system τ The impact of these three aspects on scheduling priority. In order to reasonably evaluate the scheduling priority, the scheme scores the priority of each task according to the three aspects and determines the priority according to the score ranking of each task
The priority score of (A) can be expressed as follows
Wherein, the first and the second end of the pipe are connected with each other,
indicating the arrival of the ith computation-force router r at the t time slot i And queuing the kth task to be scheduled, T, in its buffer queue 0 In order to schedule the length of the time slot,
denoted as a task to be scheduled
Delay requirement of (2), num i (t) is expressed as arriving at the ith computation-force router r at t time slot i Total task load of o t Expressed as the system time at the start of the t slot, o τ Denoted as a task to be scheduled
Time of arrival in the system, beta 1 、β 2 、β 3 Are weight factors of three types of influence factors when evaluating the priority.
And step S130, obtaining a scheduling decision through a preset comprehensive optimization target of task scheduling based on the determined schedulable task.
In this step, when selecting suitable computational service nodes for different schedulable tasks, from the perspective of a service provider, it is often desirable to ensure load balancing of the whole system while achieving efficient processing of the tasks. Therefore, in the invention, the total task processing delay and the load balance condition of the system are used as the comprehensive optimization target of task scheduling, and the optimal scheduling decision is selected for the determined schedulable task for scheduling through the comprehensive optimization target of task scheduling.
In an embodiment of the present invention, based on sub-problem two in step S120: an appropriate computational power service node is selected for the different schedulable tasks. In an edge computing system, the system scheduling decisions given by the computing force controller may be expressed as
i, j ∈ {1,2, \8230;, N }. For tasks
In the case of a composite material, for example,
represents the task as a computation of force from the entry router r i Is dispatched to a computing service node e j And occupies node e j If the scheduling behavior does not exist, then the idle virtual machine performs task processing
In an embodiment of the present invention, it is assumed that the task is not divisible, i.e. the task to be scheduled can only be processed by the virtual machine on one edge node, i.e. the virtual machine on one edge node is used for processing the task
The total task processing delay includes that the task processing delay of a certain task can be represented by the transmission delay of the task and the execution delay of the task, and for some tasks which are not scheduled in the time slot, the delay of waiting for scheduling of the task needs to be further considered. For tasks
The various delays may be specifically expressed as follows:
transmission time delay
Wherein
Representing tasks
Actual transmission data size of R ij Entry computation force router r representing the arrival of the task i With the final serving node e j The data transmission rate therebetween.
Execution delay
Wherein
Representing tasks
The calculated amount of (1), F j Is a service node e j Total task processing capacity, V j And the number of the virtual machines deployed by the service node is represented.
Waiting time delay
Wherein T is 0 Is the length of the scheduling time slot, m represents the scheduling round of waiting of the task; if the priority is higher and there is no need to wait for the next round of scheduling, m =0.
The total processing delay of the task
Can be expressed as
When i = j, the task is processed at the ingress computational effort service node directly connected to the ingress computational effort router, that is, the task does not need to be scheduled to other service nodes, and the transmission delay of the task is negligible.
Based on the selfishness of users in the actual task processing process, the variance of the virtual machine occupancy rates of all service nodes after scheduling is adopted to evaluate the load balancing condition of the network, the smaller the variance is, the more balanced the task allocation is, and the evaluation factor of the load balancing can be expressed as:
wherein, N represents N computation force service nodes in the system, and the sequence numbers are {1,2, \8230;, N }, w } respectively j Indicating that the scheduling is completed and node e j May be expressed as:
wherein, B j Representing the number of virtual machines originally in a busy state for the node,
indicates this time slot entry router r i The number of tasks with higher priority is the number of tasks which can be scheduled.
The task processing delay and the load balancing condition of the system are comprehensively considered, and from the perspective of the whole system, the comprehensive optimization target of the task can be expressed as follows:
wherein, mu 1 ，μ 2 The factor is a coefficient factor, can be determined according to a specific application scenario, and can be adjusted to different sizes to reflect the degree of influence of two factors, namely task processing delay and load balancing, on scheduling decisions.
In one embodiment of the invention, in order to obtain an efficient task scheduling strategy, commonly used solving algorithms include an integer programming algorithm, a genetic algorithm, a simulated annealing algorithm, an ant colony algorithm and other heuristic algorithms; in consideration of the long-term influence of the current scheduling decision on the system state, a related algorithm of deep reinforcement learning can be adopted, and the method can be closer to the network scene of dynamic change in reality.
And step S140, based on the obtained scheduling decision, the computation force router executes a corresponding scheduling action, schedules the schedulable task to the corresponding computation force service node for processing, and continues to queue the rest low-priority tasks for the next scheduling.
In this step, based on the scheduling decision obtained in step S130, the computational router executes a corresponding scheduling action according to the scheduling decision issued by the computational force controller, schedules the schedulable tasks with high priority to the corresponding computational service nodes for processing, and continues to queue the remaining tasks with lower priority in the queue for the next scheduling.
In the embodiment of the present invention, a task scheduling communication interaction flow of a system in each time slot is as shown in fig. 3, where a computational force controller periodically collects task arrival information of a computational force router and virtual machine occupation states of different computational force service nodes; the calculation force controller analyzes according to the collected task arrival information and the virtual machine occupation state; obtaining the scheduling priority of each task according to a task scheduling priority determining algorithm, and screening out schedulable tasks with high priority by taking the number of idle virtual machines as a limit; and obtaining a scheduling decision for the screened high-priority tasks through a dynamic task scheduling algorithm, sending the scheduling decision to each computational power router, executing corresponding scheduling actions by the computational power routers according to the decision issued by the computational power controllers, scheduling the schedulable high-priority tasks to corresponding computational power service nodes by each computational power router for processing, and continuously queuing the rest tasks with lower priorities in a queue for next scheduling.
The invention discloses a priority perception task scheduling method and a system facing a computational power network, wherein the method combines an edge computing technology and a computational power network technology to design a novel integrated architecture facing the computational power network multi-node cooperation, further defines the organization relation among functional entities under the architecture and the communication interaction flow of a computing system, considers the isomerism of computational power service nodes and the resource limitation characteristic, limits the maximum processable task number of the system in real time, distinguishes the scheduling priorities of different tasks according to the respective characteristics of the tasks to carry out flexible scheduling, comprehensively considers the task delay requirement, the workload to be scheduled of an entry computational power router, the time for the tasks to reach the system and other aspects to give the scheduling priorities of different tasks, and designs an efficient task scheduling mechanism from the perspective of a service provider to ensure the load balance of the system while meeting the low delay requirement of a user, thereby realizing the full utilization of network resources, the effective fusion of the computation and the network, the flexible sharing of the computational resources and higher task processing success rate.
Correspondingly to the method, the invention also provides a priority-aware task scheduling system for a computational power network, which includes a computer device including a processor and a memory, wherein the memory stores computer instructions, the processor is used for executing the computer instructions stored in the memory, and when the computer instructions are executed by the processor, the system realizes the steps of the method.
An embodiment of the present invention further provides a computer-readable storage medium, on which a computer program is stored, where the computer program is executed by a processor to implement the steps of the foregoing edge computing server deployment method. The computer readable storage medium may be a tangible storage medium such as Random Access Memory (RAM), memory, read Only Memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, floppy disks, hard disks, removable storage disks, CD-ROMs, or any other form of storage medium known in the art.
Those of ordinary skill in the art will appreciate that the various illustrative components, systems, and methods described in connection with the embodiments disclosed herein may be implemented as hardware, software, or combinations of both. Whether this is done in hardware or software depends upon the particular application and design constraints imposed on the solution. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present invention. When implemented in hardware, it may be, for example, an electronic circuit, an Application Specific Integrated Circuit (ASIC), suitable firmware, plug-in, function card, or the like. When implemented in software, the elements of the invention are the programs or code segments used to perform the required tasks. The program or code segments may be stored in a machine-readable medium or transmitted by a data signal carried in a carrier wave over a transmission medium or a communication link.
It is to be understood that the invention is not limited to the specific arrangements and instrumentality described above and shown in the drawings. A detailed description of known methods is omitted herein for the sake of brevity. In the above embodiments, several specific steps are described and shown as examples. However, the method processes of the present invention are not limited to the specific steps described and illustrated, and those skilled in the art can make various changes, modifications and additions, or change the order between the steps, after comprehending the spirit of the present invention.
Features that are described and/or illustrated with respect to one embodiment may be used in the same way or in a similar way in one or more other embodiments and/or in combination with or instead of the features of the other embodiments in the present invention.
The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention, and various modifications and changes may be made to the embodiment of the present invention by those skilled in the art. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims (10)
1. A priority-aware task scheduling method for a computational power network is characterized by comprising the following steps:
the method comprises the steps that a computing force controller periodically collects task arrival information of a computing force router and virtual machine occupation states of different computing force service nodes, wherein the task arrival information is information of tasks to be scheduled in a task buffer queue;
the computing force controller analyzes the collected task arrival information and the virtual machine occupation state; evaluating the scheduling priority of each task according to a preset priority determination algorithm, and determining the schedulable task in the current period by taking the number of idle virtual machines as a limit;
obtaining a scheduling decision through a comprehensive optimization target of scheduled task scheduling based on the determined schedulable task;
and based on the obtained scheduling decision, the computation force router executes corresponding scheduling action, dispatches the schedulable tasks to the corresponding computation force service nodes for processing, and continues to queue the rest low-priority tasks for waiting for next scheduling.
2. The method of claim 1, wherein a buffer queue exists on the computational router, the buffer queue being used to temporarily hold tasks that are temporarily not scheduled.
3. The method of claim 2, further comprising:
based on the limited length of the buffer queue, when the buffer queue is full, the rest tasks which cannot enter the queue are directly dispatched to the cloud center.
4. The method according to claim 1, wherein the predetermined priority determination algorithm determines the scheduling priority of the task to be scheduled in the system based on the priority scoring of the task to be scheduled by integrating the delay requirements of each task, the total task load reaching the computation force router and the time of the task reaching the system; the priority score is expressed as:
wherein the content of the first and second substances,
indicating the arrival of the ith computation-force router r at the t time slot i And queued the kth task to be scheduled, T, in its buffer queue 0 In order to schedule the length of the time slot,
denoted as a task to be scheduled
Delay requirement of (2), num i (t) is the total task load or buffer queue length to reach the ith computation router at time slot t, o t Indicating the system time at the start of the t-slot, o τ Denoted as a task to be scheduled
Time of arrival in the system, beta 1 、β 2 、β 3 The weight factors of the three types of influence factors when evaluating the priority.
5. The method according to claim 1, wherein the schedulable tasks in the current period are determined by taking the number of idle virtual machines as a limit, and the total number of the final scheduling tasks in the system is not more than the number of idle virtual machines.
6. The method of claim 1, wherein the comprehensive optimization goal of the predetermined task scheduling is total task processing delay and load balancing of the system; the scheduling decision is to select proper computational power service nodes for different schedulable tasks;
the total task processing time delay comprises the transmission time delay of the task and the execution time delay of the task, and for the task which is not scheduled in the scheduling time slot, the total task processing time delay also comprises the time delay of waiting for scheduling of the task;
for the k-th task to be scheduled which arrives at the ith computation force router at the t time slot and is queued in the buffer queue of the ith computation force router
The various delays are represented as follows:
the transmission delay is expressed as
Wherein the content of the first and second substances,
representing tasks
Actual transmission data size of R ij Entry computation force router r representing the arrival of the current task i With the final serving node e j The data transmission rate therebetween;
the execution delay is expressed as
Wherein, the first and the second end of the pipe are connected with each other,
representing tasks
Size of calculated amount of (D), F j Is a service node e j Total task processing capacity, V j Representing the number of virtual machines deployed by the service node;
the latency is expressed as
Wherein, T 0 The length of a scheduling time slot is m, and the m represents the scheduling turn of waiting of the current task; if the priority is higher and does not need to wait for the next round of scheduling, m =0;
the total task processing delay of the current task is represented as:
when i = j, it means that the current task is processed at the computation force service node directly connected to the computation force router without being scheduled to other service nodes.
7. The method of claim 6, wherein the computing power service nodes directly connected to the computing power routers are ingress service nodes, and each computing power router has a computing power service node directly connected to it.
8. The method of claim 6, further comprising using the variance of the virtual machine occupancy of each computing power service node after scheduling is completed to evaluate the load balancing of the network.
9. A priority-aware task scheduling system for computational power networks, comprising a processor and a memory, wherein the memory has stored therein computer instructions for executing the computer instructions stored in the memory, which when executed by the processor, implements the steps of the method according to any one of claims 1 to 8.
10. A computer-readable storage medium, on which a computer program is stored which, when being executed by a processor, carries out the steps of a method according to any one of claims 1 to 8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210929252.7A CN115421901A (en) | 2022-08-03 | 2022-08-03 | Priority perception task scheduling method and system for computational power network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210929252.7A CN115421901A (en) | 2022-08-03 | 2022-08-03 | Priority perception task scheduling method and system for computational power network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115421901A true CN115421901A (en) | 2022-12-02 |
Family
ID=84197285
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210929252.7A Pending CN115421901A (en) | 2022-08-03 | 2022-08-03 | Priority perception task scheduling method and system for computational power network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115421901A (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116089046A (en) * | 2023-01-31 | 2023-05-09 | 安徽航天联志科技有限公司 | Scheduling method, device, equipment and medium based on software-defined computing network |
| CN116136799A (en) * | 2023-04-14 | 2023-05-19 | 亚信科技（中国）有限公司 | Computing power dispatching management side device and method, computing power providing side device and method |
| CN116302568A (en) * | 2023-05-17 | 2023-06-23 | 算力互联(北京)科技有限公司 | Computing power resource scheduling method and system, scheduling center and data center |
| CN116737178A (en) * | 2023-08-10 | 2023-09-12 | 北京万界数据科技有限责任公司 | Training task arrangement method and system |
| CN116846818A (en) * | 2023-09-01 | 2023-10-03 | 北京邮电大学 | Method, system, device and storage medium for dispatching traffic of computing power network |
| CN117478529A (en) * | 2023-12-27 | 2024-01-30 | 环球数科集团有限公司 | Distributed computing power sensing and scheduling system based on AIGC |
-
2022
- 2022-08-03 CN CN202210929252.7A patent/CN115421901A/en active Pending
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116089046A (en) * | 2023-01-31 | 2023-05-09 | 安徽航天联志科技有限公司 | Scheduling method, device, equipment and medium based on software-defined computing network |
| CN116136799A (en) * | 2023-04-14 | 2023-05-19 | 亚信科技（中国）有限公司 | Computing power dispatching management side device and method, computing power providing side device and method |
| CN116302568A (en) * | 2023-05-17 | 2023-06-23 | 算力互联(北京)科技有限公司 | Computing power resource scheduling method and system, scheduling center and data center |
| CN116737178A (en) * | 2023-08-10 | 2023-09-12 | 北京万界数据科技有限责任公司 | Training task arrangement method and system |
| CN116737178B (en) * | 2023-08-10 | 2023-10-20 | 北京万界数据科技有限责任公司 | Training task arrangement method and system |
| CN116846818A (en) * | 2023-09-01 | 2023-10-03 | 北京邮电大学 | Method, system, device and storage medium for dispatching traffic of computing power network |
| CN116846818B (en) * | 2023-09-01 | 2023-12-01 | 北京邮电大学 | Method, system, device and storage medium for dispatching traffic of computing power network |
| CN117478529A (en) * | 2023-12-27 | 2024-01-30 | 环球数科集团有限公司 | Distributed computing power sensing and scheduling system based on AIGC |
| CN117478529B (en) * | 2023-12-27 | 2024-03-12 | 环球数科集团有限公司 | Distributed computing power sensing and scheduling system based on AIGC |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN115421901A (en) | Priority perception task scheduling method and system for computational power network | |
| Beraldi et al. | Distributed load balancing for heterogeneous fog computing infrastructures in smart cities | |
| Qian et al. | Survey on reinforcement learning applications in communication networks | |
| Aujla et al. | An ensembled scheme for QoS-aware traffic flow management in software defined networks | |
| Zhou et al. | Learning from peers: Deep transfer reinforcement learning for joint radio and cache resource allocation in 5G RAN slicing | |
| US11601876B2 (en) | Method for controlling the admission of slices into a virtualized telecommunication network and the congestion likely to be generated between services instantiated on said slices | |
| Sun et al. | Enhancing the user experience in vehicular edge computing networks: An adaptive resource allocation approach | |
| Chuprikov et al. | Priority queueing with multiple packet characteristics | |
| Al-Turjman et al. | SAHCI: scheduling approach for heterogeneous content-centric IoT applications | |
| Ali et al. | Admission control-based multichannel data broadcasting for real-time multi-item queries | |
| Abdollahi et al. | Flow-aware forwarding in SDN datacenters using a knapsack-PSO-based solution | |
| Li et al. | Software-defined vehicular networks with caching and computing for delay-tolerant data traffic | |
| Rashid et al. | Integrated sized-based buffer management policy for resource-constrained delay tolerant network | |
| Lan et al. | Throughput-optimal H-QMW scheduling for hybrid wireless networks with persistent and dynamic flows | |
| Ullah et al. | Statistical multipath queue-wise preemption routing for zigbee-based WSN | |
| CN106911593A (en) | A kind of industrial control network array dispatching method based on SDN frameworks | |
| Jagabathula et al. | Fair scheduling in networks through packet election | |
| CN109547345B (en) | Software-defined airborne network system and content-driven routing method | |
| Zhang et al. | Vehicular multi-slice optimization in 5G: Dynamic preference policy using reinforcement learning | |
| Rezaee et al. | A fuzzy algorithm for adaptive multilevel queue management with QoS feedback | |
| Guo et al. | Joint wireless resource allocation and service function chaining scheduling for Tactile Internet | |
| Sedaghat et al. | R2T-DSDN: reliable real-time distributed controller-based SDN | |
| Khawam et al. | Opportunistic weighted fair queueing | |
| Zhu et al. | Minimizing Age-of-Information with Joint Transmission and Computing Scheduling in Mobile Edge Computing | |
| CN116133049B (en) | Cloud edge end collaborative MEC task unloading strategy based on DRL and safety |
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 |