| CPC G05B 19/41885 (2013.01) [G05B 19/4185 (2013.01)] | 3 Claims |
|
1. A task-level cooperative optimization dispatching method supporting multi-cloud disconnection disaster recovery, comprising the following steps:
assigning, by a manufacturing node, a task to other matched manufacturing nodes through an end-to-end communication channel not dependent on a service;
conducting cooperative optimization dispatching for the task between the manufacturing node and other manufacturing nodes by a cooperative dispatch method;
synchronizing task processing results of the manufacturing node and other manufacturing nodes among the manufacturing nodes by a locality mechanism,
wherein the cooperative dispatch method calculates a resource-limited minimum task execution average time in a cooperative computing system composed of a plurality of manufacturing nodes to cooperatively dispatch the task to improve the overall revenue of the cooperative computing system, comprising the following steps:
1) dividing tasks in the cooperative computing system CS into class |M|; representing a task set as Task; for the tasks in class m, m∈|M|, and processing the tasks only in the task set Taskm∈Task where resource Ri exists; for the current manufacturing node, an arrival rate Aim(t) of a request for the tasks in class m to other manufacturing nodes at time t satisfies:
0≤Aim(t)≤Ammax*τ{Ti∈Taskm}
wherein Ammax represents a maximum arrival rate of the tasks in class m, i represents an ith task of the tasks in class m, τ{Ti∈Taskm} is a flag bit for representing whether the task Ti that can be processed by the resource Ri is in the task set Taskm, wherein:
 the manufacturing node processes multiple tasks at the same time, the tasks in class m at time t are stored in a queue Qim(t), and all task queues {Qim(t), ri∈Ri, Taskm∈Task} constitute the following queue matrix Q(t):
 ri represents a resource in resource set Ri, and QI|M|(t) represents a last task of the processed tasks in class m with a queue waiting number of |M|; Qim(t) represents a task in class m processed by the resource ri, i=1 . . . I;
the number of requests of the task in class m received at time t is defined as aim(t), which satisfies:
0≤aim(t)≤Aim(t)*Δt,
wherein Δt represents a time interval, which is a constant;
the task execution average time is as follows:
 wherein α is an empirical constant, E{ } represents an expectation, which indicates the expectation of the task execution average time aim;
2) assigning, by the cooperative computing system CS, resources required by the tasks to process the tasks, and assigning the resource CSi corresponding to the ith task to the resource requested by the task in class m, rim(t):
Σmrim(t)≤Ri,
updating a task processing queue by the manufacturing node through the following model:
Qim(t+1)=max(Qim(t)−rim(t)*∂m),
wherein ∂m represents the number of the tasks in class m that can be processed by unit resource in a set cycle;
3) constructing a comprehensive benefit model according to the task execution average time aim and the resource rim(t), and obtaining a combination of minimum execution average time aim and resource rim(t) by comprehensive benefit maximization of the collaborative computing system CS to achieve cooperative optimization dispatching,
wherein:
in step 3), achieving cooperative optimization dispatching by comprehensive benefit maximization of the collaborative computing system CS comprises the following steps:
(1) when the task of the manufacturing node is received by other manufacturing nodes, the manufacturing node obtains the following benefits:
∫1(aim)=log(1+βm*aim),
wherein ∫1(aim) represents the first part of benefit, βm is a request revenue constant of the task in class m and aim is the task execution average time;
(2) the number nm(t) of requests of the task in class m processed by the manufacturing node at time t is:
nm(t)=Σirim(t)*∂m,
∂m represents the number of the tasks in class m that can be processed by unit resource in a set cycle;
corresponding revenue is:
∫2(nm)=θm*nm,
wherein nm is a time expectation of nm(t), and θm is a corresponding revenue constant; ∫2(nm) represents the second part of benefit, which depends on the number of served requests;
(3) the comprehensive benefit model is:
B=Σi,m∫1(aim)+Σm∫2(nm),
wherein i represents the ith task of the tasks in class m; and
(4) to maximize the comprehensive benefit to achieve cooperative optimization, aim obtained is the minimum execution average time when B reaches a maximum value,
wherein:
synchronizing task processing results of the manufacturing node and other manufacturing nodes among the manufacturing nodes by the locality mechanism comprises the following steps:
when the manufacturing node receives the tasks of other nodes, with the first part of benefit ∫1(aim) being nonlinear, adding a transform constant variable yim(t) and a constant variable transposition queue {Yim(t), ∀i, m} to convert task optimization dispatching into a linear problem:
Φ(Ω(t))=Σi,mE{Qim(t)*wm*aim(t)−Yim(t)*∂m*aim(t)|(Ω(t)},
wherein wm and ∂m are constant variables, Φ(Ω(t)) represents a maximum comprehensive benefit value, Ω(t) represents a difference between a current comprehensive benefit and an average comprehensive benefit, E{ } represents an expectation which here represents an expected value of benefit, and aim is the task execution average time;
substituting the combination of the execution average time aim and the resource rim(t) obtained by comprehensive benefit maximization of the cooperative computing system CS in step 3) into Φ(Ω(t)), and minimizing Φ(Ω(t)) to obtain an optimized task dispatching strategy;
wherein
 when a real queue Qim(t) is not longer than the constant variable transposition queue, receiving, by the manufacturing node, all requests of the tasks in class m that reach at time t and executing the requested tasks, otherwise, waiting.
|