CN107633125B - Simulation system parallelism identification method based on weighted directed graph - Google Patents
Simulation system parallelism identification method based on weighted directed graph Download PDFInfo
- Publication number
- CN107633125B CN107633125B CN201710826327.8A CN201710826327A CN107633125B CN 107633125 B CN107633125 B CN 107633125B CN 201710826327 A CN201710826327 A CN 201710826327A CN 107633125 B CN107633125 B CN 107633125B
- Authority
- CN
- China
- Prior art keywords
- simulation component
- directed graph
- directed
- component model
- weighted
- 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
Abstract
The invention discloses a simulation system parallelism identification method based on a weighted directed graph, which comprises the following steps: s1, constructing a directed graph which takes the simulation component models as nodes and takes the connection between the simulation component models as directed edges according to the connection relation of the simulation component models in the simulation system and the time sequence relation of the generated state transition; s2, calculating the calculation cost of each node and the communication cost of the directed edge in the directed graph, and constructing a weighted directed graph which comprises the weight taking the calculation cost of the node as the node and the weight taking the communication cost of the directed edge as the directed edge; and S3, identifying all parallel branches in the weighted directed graph, pruning the weighted directed graph according to the weights of the directed edges, taking the result of performing fusion operation on all the weights contained in the parallel branches in the pruned weighted directed graph as the weights of the parallel branches, grouping the parallel branches according to the weights of the parallel branches and distributing computing resources for each group. The invention can balance the calculation overhead.
Description
Technical Field
The invention relates to the technical field of simulation. And more particularly, to a simulation system parallelism identification method based on a weighted directed graph.
Background
The simulation system is used for generating corresponding state transition and output tracks on the premise of giving system description (models), initial conditions and input time sequences so as to simulate and observe objective world systems (which can be natural systems, social systems and engineering systems). A simulation system is generally described by being divided into a plurality of parts according to the composition of the system, and describing the parts by using different description methods. I.e. a simulation system typically comprises a plurality of heterogeneous submodels. At present, the popular practice is to encapsulate heterogeneous submodels into normalized simulation component models (the simulation component models include atomic component models and composite component models which are not separable, and the composite component models are formed by connecting atomic component models and/or composite component models with smaller granularity through interfaces of the models and the models), and the rapid construction of a complex simulation system is realized through the combination of the simulation component models.
The operation of the simulation system is time-constrained, and the state of each simulation component model is transferred along with the time advance under the drive of the input. The parallel operation of the simulation system can respectively calculate the state change of each simulation component model, and then the state transition between the simulation component models is ensured to accord with objective causal relationship through information interaction with time sequence. However, the parallel operation of the conventional simulation system is coarse-grained, that is, a coarse-grained composite component model is allocated to different computing nodes to be executed in parallel. However, for the latest multi-core or many-core computing resources, such parallelization is far from taking advantage of the computing resources, and a finer-grained simulation component model needs to be allocated to a CPU or even a core for parallel execution.
The current parallelization technology cannot provide decision support for resource allocation of fine-grained parallel operation, wherein the computational overhead of a simulation component model in a time period is determined by the computational complexity of state transition of the simulation component model and the time sequence relation between the simulation component models (internal or external information for causing the state transition of the model is required to drive the computation of the state transition of the model).
Therefore, it is required to provide a simulation system parallelism identification method based on weighted directed graph, which can realize reasonable grouping between simulation component models with fine granularity, so that the computation overhead of each group can be balanced in a time period
Disclosure of Invention
The invention aims to provide a simulation system parallelism identification method based on a weighted directed graph.
In order to achieve the purpose, the invention adopts the following technical scheme:
a simulation system parallelism identification method based on a weighted directed graph comprises the following steps:
s1, obtaining the connection relation of each simulation component model according to the information interaction relation of each simulation component model, and constructing a directed graph which takes the simulation component models as nodes and takes the connection between the simulation component models as directed edges according to the connection relation of each simulation component model in the simulation system and the time sequence relation of state transition;
s2, calculating the calculation cost of each node and the communication cost of the directed edge in the directed graph, and constructing a weighted directed graph which comprises the weight taking the calculation cost of the node as the node and the weight taking the communication cost of the directed edge as the directed edge;
and S3, identifying all parallel branches in the weighted directed graph, pruning the weighted directed graph according to the weight of the directed edge, performing fusion operation on all weights contained in the parallel branches in the pruned weighted directed graph, taking the fusion operation result as the weight of the parallel branches, grouping the parallel branches according to the weight of the parallel branches, and distributing computing resources for each parallel branch group.
Preferably, the step S1 further includes:
s1.1, obtaining the connection relation of each simulation component model according to the information interaction relation of each simulation component model, and taking the simulation component model as a node;
s1.2, taking a simulation component model which generates state transition at first as a starting point;
s1.3, generating a directed edge from the starting point to the end point by taking the simulation component model connected with the simulation component model representing the starting point as the end point;
s1.4, taking a simulation component model representing an end point as a starting point;
s1.5, judging whether all simulation component models are taken as a starting point: if not, the step S1.3 is carried out; and if so, constructing a directed graph containing the nodes and the directed edges.
Preferably, the step S2 further includes:
s2.1, carrying out weighted averaging on the state transition calculation expense driven by external information generated by connection of directed edge representatives with the simulation component model as an end point and the state transition calculation expense driven by internal information of the simulation component model to obtain the calculation expense of the nodes represented by the simulation component model, and normalizing the calculation expense of each node to obtain the weight of each node;
s2.2, counting information interaction generated by connection of information interaction frequency and information interaction data quantity to obtain communication overhead of directed edges represented by connection, and normalizing communication overhead of the directed edges to obtain weights of the directed edges;
and S2.3, constructing a weighted directed graph containing the weight of the node and the weight of the directed edge.
Preferably, the step S3 further includes:
s3.1, identifying all parallel branches in the weighted directed graph according to the time sequence of state transition generation of a simulation component model serving as a node in the weighted directed graph and connection serving as a directed edge;
s3.2, pruning the weighted directed graph according to the weight of the directed edge, performing fusion operation on all weights contained in the parallel branches in the pruned weighted directed graph, taking the fusion operation result as the weight of the parallel branches, independently grouping the parallel branches with large weight, and combining and grouping the parallel branches with small weight so as to enable the weight of each parallel branch group to be approximate;
and S3.3, distributing computing resources for each parallel branch group.
The invention has the following beneficial effects:
the technical scheme of the invention is based on the traditional coarse-grained identification method that a simulation system based on a composite component model can run in parallel, and takes the calculation overhead of different simulation component models and the communication overhead between the simulation component models in a time period into consideration to construct a weighted directed graph, so that the reasonable grouping of the simulation component models with fine granularity is realized, the calculation overhead of each group in a time period can be balanced, the communication overhead of each group is small (the simulation component models with frequent information interaction and large information interaction data volume are divided into one group as much as possible), and therefore, under the condition of the latest multi-core or multi-core calculation resources, decision support can be provided for distributing the simulation component models with fine granularity to a CPU (Central processing Unit) and even a core for parallel execution.
Drawings
The following describes embodiments of the present invention in further detail with reference to the accompanying drawings;
fig. 1 shows a description diagram of a simulation system.
FIG. 2 shows a flow chart of a simulation system parallelism recognition method based on a weighted directed graph.
Fig. 3 shows a schematic diagram of a weighted directed graph.
Fig. 4 shows a schematic diagram of a weighted directed graph after pruning.
Detailed Description
In order to more clearly illustrate the invention, the invention is further described below with reference to preferred embodiments and the accompanying drawings. Similar parts in the figures are denoted by the same reference numerals. It is to be understood by persons skilled in the art that the following detailed description is illustrative and not restrictive, and is not to be taken as limiting the scope of the invention.
As shown in fig. 2, the simulation system parallelism identification method based on the weighted directed graph provided in this embodiment includes:
s1, obtaining the connection relation of each simulation component model according to the information interaction relation of each simulation component model, and constructing a directed graph which comprises the connection between simulation component models serving as nodes and the connection between the simulation component models serving as directed edges according to the connection relation of each simulation component model in the simulation system and the time sequence relation of state transition, wherein the directed graph is generally provided with a closed loop, and output signals of each simulation component model are transmitted and converted through a plurality of simulation component models and then are probably transmitted back to the input interface of the simulation component model;
s2, calculating the calculation cost of each node and the communication cost of the directed edge in the directed graph, and constructing a weighted directed graph which comprises the weight taking the calculation cost of the node as the node and the weight taking the communication cost of the directed edge as the directed edge;
and S3, identifying all parallel branches in the weighted directed graph, pruning the weighted directed graph according to the weight of the directed edge, performing fusion operation on all weights contained in the parallel branches in the pruned weighted directed graph, taking the fusion operation result as the weight of the parallel branches, grouping the parallel branches according to the weight of the parallel branches, and distributing computing resources for each parallel branch group.
In a specific implementation, step S1 further includes:
s1.1, obtaining the connection relation of each simulation component model according to the information interaction relation of each simulation component model, and taking the simulation component model as a node;
s1.2, taking a simulation component model which generates state transition at first as a starting point;
s1.3, generating a directed edge from the starting point to the end point by taking the simulation component model connected with the simulation component model representing the starting point as the end point;
s1.4, taking a simulation component model representing an end point as a starting point;
s1.5, judging whether all simulation component models are taken as a starting point: if not, the step S1.3 is carried out; and if so, constructing a directed graph containing the nodes and the directed edges.
In a specific implementation, step S2 further includes:
s2.1, carrying out weighted averaging on the state transition calculation expense driven by external information generated by connection of directed edge representatives with the simulation component model as an end point and the state transition calculation expense driven by internal information of the simulation component model to obtain the calculation expense of the nodes represented by the simulation component model, and normalizing the calculation expense of each node to obtain the weight of each node;
s2.2, counting the information interaction generated by connection to obtain the communication overhead of the directed edges represented by the connection, wherein the counting comprises the statistics of the information interaction frequency between the nodes connected by each directed edge and the data volume of single information interaction; then, respectively normalizing the information interaction frequency overhead and the single information interaction data volume overhead of each directed edge to obtain the weight of each directed edge;
and S2.3, constructing a weighted directed graph containing the weight of the node and the weight of the directed edge.
In a specific implementation, step S3 further includes:
s3.1, identifying all parallel branches in the weighted directed graph according to the time sequence of state transition generation of a simulation component model serving as a node in the weighted directed graph and connection serving as a directed edge;
s3.2, because the state transition is generated among the simulation component models according to the causal relationship and the time sequence, the appropriate grouping combination can be carried out to solve the problem that different groups wait for each other(resulting in wasted idle resources); consider a weighted directed graph as shown in FIG. 3, where each node computes an overhead weight of ciWherein i is equal to {1, …,11} is node label, and information interaction frequency overhead is wijThe subscript indicates that information interaction is carried out from node i to node j, and the data volume overhead of single information interaction is qij. Considering the information interaction frequency and the data volume overhead of each directed edge, set w75*q75If the weighted value is far smaller than the weighted value of the rest edges, the edges c7-c5 are pruned, and a weighted directed graph shown in FIG. 4 is obtained after pruning.
Calculating the weighted directed graph after pruning according to the weight on each forward path according to the following formula
The reduced overhead value for each forward path is obtained. And the nodes of the parallel branches with large overhead values are separately grouped, and the parallel branches with small weights are combined into one group, so that the weights of all the parallel branch groups are balanced. For c as in FIG. 47-c8-c6、c4-c5-c6And c7-c8-c9In case of c7-c8-c6If the comprehensive overhead is large, c is added4-c5Are divided into a group c7-c8-c6Are grouped into groups, in which case if c8-c9If the overhead is also large, it is divided into a group separately, otherwise it is incorporated into c7-c8-c6Group (d); if c is7-c8-c6The overhead is not large, and c4-c5-c6And c7-c8-c9If the expenses are large, the expenses can be randomly divided into one group; if c is7-c8-c6The overhead is not large, and c4-c5-c6And c7-c8-c9One group of the overhead is large, the other group of the overhead is not large, c is added7-c8-c6A group with small overhead is divided; if the overhead of the above three groups is not large, the method can be divided intoThe same group. The above evaluation of the overhead needs to be considered according to the available computing resources.
And S3.3, computing resources are allocated to each parallel branch group, namely, the simulation component model with large communication overhead (namely frequent state transition and large information interaction data volume) and large computing overhead is allocated with single resources.
The simulation system parallelism identification method based on the weighted directed graph provided by the embodiment is directed at the situation that the existing coarse-grained simulation system runs in parallel without considering the computation overhead of different simulation component models and the communication overhead among the simulation component models in a time period, and provides the identification method capable of executing the fine-grained simulation component models in parallel, so that the grouping of the fine-grained simulation component models is realized, the computation overhead of each group in the time period can be balanced, the communication overhead among the groups is small (the simulation component models with frequent information interaction and large information interaction data amount are divided into one group as much as possible), and therefore under the condition of the latest multi-core or multi-core computing resources, decision support can be provided for distributing the fine-grained simulation component models to a CPU and even a core for parallel execution.
The simulation system parallelism identification method based on the weighted directed graph provided by the embodiment is further explained by taking a multidisciplinary virtual prototype simulation system constructed based on a high-level modeling theory as an example:
the method comprises the following steps of firstly, obtaining the connection relation of each simulation component model according to the information interaction relation of each simulation component model, and constructing a directed graph which takes the simulation component models as nodes and the connection between the simulation component models as directed edges according to the connection relation of each simulation component model in a simulation system and the time sequence relation of state transition generation:
atomic component model: the atomic component model is composed of a parameter interface, an initialization interface, a data input and output interface, an event input and output interface, states and conversion thereof, and a time base.
EM＝<Para,Init,iPd,iPe,oPd,oPe,STATE,StatTF,T>
Wherein the content of the first and second substances,
para: a parameter interface;
init: initializing an interface;
iPd: a data input interface;
iPe: an event input interface;
oPd: a data output interface;
oPe: an event output interface;
state: presenting a state on the interface;
StatTF: state conversion;
a composite component model: the composite component model is composed of a parameter interface, an initialization interface, a data input and output interface, an event input and output interface, model internal connection, an interaction scenario, an event stream and a time base.
CM＝<Para,Init,iPd,iPe,oPd,oPe,ID,{Mx|x∈ID∪{CM}},CPLs,SITUA,EvntFL,T>
Wherein the content of the first and second substances,
para: a parameter interface;
init: initializing an interface;
iPd: a data input interface;
iPe: an event input interface;
oPd: a data output interface;
oPe: an event output interface;
ID: a set of model indices;
mx: the CM or a subcomponent model contained in the CM;
CPLs: a set of internal connections of the composite model;
SITUA: a set of interaction scenarios;
EventFL: event streams inside the composite model;
t: and (4) time base.
Taking a simulation component model (certain EM) which generates state transition at first as a starting point as a source, and generating an initial starting point of a directed graph; along the connection relation (CPLS) of the simulation component model, corresponding the connection where the output interface (oPd or oPe) is located to a directed edge taking the starting point as the starting end; taking the simulation component model (corresponding EM) connected with the orientation as a tail end, and generating a directed edge; and taking the simulation component model at the tail end as a starting point, circularly executing the process, traversing all the simulation component models, and constructing a complete directed graph with a closed loop normally (because the output signals of all the simulation component models are transmitted and converted by a plurality of simulation component models and are possibly transmitted back to the input interface of the simulation component model finally).
Secondly, calculating the calculation cost of each node and the communication cost of the directed edge in the directed graph, and constructing a weighted directed graph which comprises the weight taking the calculation cost of the node as the node and the weight taking the communication cost of the directed edge as the directed edge:
the calculation cost of the simulation component models represented by each node in the directed graph is considered, and mainly includes the calculation cost of external information-driven state transition caused by the connection relation between the simulation component models represented by the directed edges connected with the tail end and the calculation cost of state transition (calculation cost of StatTF) driven by the internal information of the simulation component models represented by the node.
iPe state transitions driven by external information when not empty; iPe, when empty, it is the state transition driven by internal information), and the two types of overhead are weighted and averaged (after normalization of each node) to obtain the weight of the node; in addition, considering the communication overhead of the connection relationship between the simulation component models represented by the directional edges, the weights of the directional edges are generally obtained by counting (normalizing) the information interaction (oPd and oPe generated when the simulation is executed) generated by each connection.
Thirdly, identifying all parallel branches in the weighted directed graph, pruning the original network according to the edge weight obtained by calculating the communication overhead, carrying out fusion operation on all weights contained in the parallel branches in the pruned network, taking the result as the weight of the parallel branches, grouping the parallel branches according to the weight of the parallel branches, and allocating calculation resources for each parallel branch group:
and traversing the weighted directed graph, and identifying all parallel branches as a maximum parallelable grouping scheme. Due to the fact that the state transition is generated between the simulation component models according to the causal relationship and the time sequence, the appropriate grouping combination can be carried out, and the problem that different groups wait for each other (the resource is idle and wasted) is solved. The method comprises the steps of pruning the weighted directed graph by the weight of the directed edge, carrying out fusion operation on all weights contained in parallel branches in the pruned weighted directed graph, taking branches with larger results to carry out independent grouping, taking branches with smaller results to carry out grouping and merging (ensuring that the sum of products of the grouped and grouped products is basically equivalent to the performance of the independent grouping), and finally allocating computing resources for each parallel branch group, namely allocating the individual resources to a simulation component model with large communication overhead (namely frequent state transition and large data interaction data amount) and large computing overhead.
It should be understood that the above-mentioned embodiments of the present invention are only examples for clearly illustrating the present invention, and are not intended to limit the embodiments of the present invention, and it will be obvious to those skilled in the art that other variations or modifications may be made on the basis of the above description, and all embodiments may not be exhaustive, and all obvious variations or modifications may be included within the scope of the present invention.
Claims (3)
1. A simulation system parallelism identification method based on a weighted directed graph is characterized by comprising the following steps:
s1, obtaining the connection relation of each simulation component model according to the information interaction relation of each simulation component model, and constructing a directed graph which takes the simulation component models as nodes and takes the connection between the simulation component models as directed edges according to the connection relation of each simulation component model in the simulation system and the time sequence relation of state transition;
s2, calculating the calculation cost of each node and the communication cost of the directed edge in the directed graph, and constructing a weighted directed graph which comprises the weight taking the calculation cost of the node as the node and the weight taking the communication cost of the directed edge as the directed edge;
s3, identifying all parallel branches in the weighted directed graph, pruning the weighted directed graph according to the weights of the directed edges, performing fusion operation on all weights contained in the parallel branches in the pruned weighted directed graph, taking the fusion operation result as the weight of the parallel branches, grouping the parallel branches according to the weight of the parallel branches, and distributing computing resources for each parallel branch group;
step S3 further includes:
s3.1, identifying all parallel branches in the weighted directed graph according to the time sequence of state transition generation of a simulation component model serving as a node in the weighted directed graph and connection serving as a directed edge;
s3.2, pruning the weighted directed graph according to the weight of the directed edge, performing fusion operation on all weights contained in the parallel branches in the pruned weighted directed graph, taking the fusion operation result as the weight of the parallel branches, independently grouping the parallel branches with large weight, and combining and grouping the parallel branches with small weight so as to enable the weight of each parallel branch group to be approximate;
and S3.3, distributing computing resources for each parallel branch group.
2. The simulation system parallelism recognition method based on the weighted directed graph according to claim 1, wherein the step S1 further comprises:
s1.1, obtaining the connection relation of each simulation component model according to the information interaction relation of each simulation component model, and taking the simulation component model as a node;
s1.2, taking a simulation component model which generates state transition at first as a starting point;
s1.3, generating a directed edge from the starting point to the end point by taking the simulation component model connected with the simulation component model representing the starting point as the end point;
s1.4, taking a simulation component model representing an end point as a starting point;
s1.5, judging whether all simulation component models are taken as a starting point: if not, the step S1.3 is carried out; and if so, constructing a directed graph containing the nodes and the directed edges.
3. The simulation system parallelism recognition method based on the weighted directed graph according to claim 2, wherein the step S2 further comprises:
s2.1, carrying out weighted averaging on the state transition calculation expense driven by external information generated by connection of directed edge representatives with the simulation component model as an end point and the state transition calculation expense driven by internal information of the simulation component model to obtain the calculation expense of the nodes represented by the simulation component model, and normalizing the calculation expense of each node to obtain the weight of each node;
s2.2, counting information interaction generated by connection of information interaction frequency and information interaction data quantity to obtain communication overhead of directed edges represented by connection, and normalizing communication overhead of the directed edges to obtain weights of the directed edges;
and S2.3, constructing a weighted directed graph containing the weight of the node and the weight of the directed edge.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710826327.8A CN107633125B (en) | 2017-09-14 | 2017-09-14 | Simulation system parallelism identification method based on weighted directed graph |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710826327.8A CN107633125B (en) | 2017-09-14 | 2017-09-14 | Simulation system parallelism identification method based on weighted directed graph |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107633125A CN107633125A (en) | 2018-01-26 |
| CN107633125B true CN107633125B (en) | 2021-08-31 |
Family
ID=61101923
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710826327.8A Active CN107633125B (en) | 2017-09-14 | 2017-09-14 | Simulation system parallelism identification method based on weighted directed graph |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107633125B (en) |
Families Citing this family (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108563984B (en) * | 2018-03-02 | 2021-12-03 | 山东科技大学 | Automatic identification and understanding method of flow model diagram |
| CN110610227B (en) * | 2018-06-15 | 2022-07-26 | 赛灵思电子科技(北京)有限公司 | Artificial neural network adjusting method and neural network computing platform |
| CN110865875B (en) * | 2018-08-27 | 2023-04-11 | 阿里巴巴集团控股有限公司 | DAG task relationship graph processing method and device and electronic equipment |
| CN110147273A (en) * | 2019-05-17 | 2019-08-20 | 深圳前海微众银行股份有限公司 | A kind of task executing method and device |
| CN110366205B (en) * | 2019-08-05 | 2020-10-30 | 北京邮电大学 | Method and device for selecting initial source node in mobile opportunity network traffic unloading |
| CN111046575A (en) * | 2019-12-23 | 2020-04-21 | 中国航空工业集团公司沈阳飞机设计研究所 | Method and system for ensuring simulation consistency |
| CN111245662B (en) * | 2020-03-09 | 2023-01-24 | 杭州迪普科技股份有限公司 | Method and device for displaying network topology |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103605560A (en) * | 2013-11-25 | 2014-02-26 | 国家电网公司 | Cascading failure parallel simulation method for relay protection and safety automata |
| CN106096159A (en) * | 2016-06-20 | 2016-11-09 | 华北电力大学（保定） | Distributed system behavior simulation under a kind of cloud platform analyzes the implementation method of system |
| CN106682369A (en) * | 2017-02-27 | 2017-05-17 | 常州英集动力科技有限公司 | Heating pipe network hydraulic simulation model identification correction method and system, method of operation |
Family Cites Families (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| FR3009406B1 (en) * | 2013-08-02 | 2016-12-23 | Commissariat Energie Atomique | METHOD FOR DETERMINING A RADIUS TRACE BETWEEN TWO POINTS MODELING THE PROPAGATION OF AN ULTRASONIC WAVE CROSSING A STRUCTURE WITH DIFFRACTIVE INTERFACES |
| CN103885839B (en) * | 2014-04-06 | 2017-02-22 | 孙凌宇 | Cloud computing task scheduling method based on multilevel division method and empowerment directed hypergraphs |
| CN104391889A (en) * | 2014-11-11 | 2015-03-04 | 西安交通大学 | Method for discovering community structure oriented to directed-weighting network |
| CN105786626B (en) * | 2016-04-11 | 2019-02-19 | 南京邮电大学 | The Storm platform thread distribution method divided based on K |
| CN106407007B (en) * | 2016-08-31 | 2020-06-12 | 上海交通大学 | Cloud resource configuration optimization method for elastic analysis process |
-
2017
- 2017-09-14 CN CN201710826327.8A patent/CN107633125B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103605560A (en) * | 2013-11-25 | 2014-02-26 | 国家电网公司 | Cascading failure parallel simulation method for relay protection and safety automata |
| CN106096159A (en) * | 2016-06-20 | 2016-11-09 | 华北电力大学（保定） | Distributed system behavior simulation under a kind of cloud platform analyzes the implementation method of system |
| CN106682369A (en) * | 2017-02-27 | 2017-05-17 | 常州英集动力科技有限公司 | Heating pipe network hydraulic simulation model identification correction method and system, method of operation |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107633125A (en) | 2018-01-26 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107633125B (en) | Simulation system parallelism identification method based on weighted directed graph | |
| US20220391771A1 (en) | Method, apparatus, and computer device and storage medium for distributed training of machine learning model | |
| Ribeiro et al. | Mlaas: Machine learning as a service | |
| US8601458B2 (en) | Profile-driven data stream processing | |
| CN103488775B (en) | A kind of calculating system processed for big data and computational methods | |
| CN114756383A (en) | Distributed computing method, system, device and storage medium | |
| US9053067B2 (en) | Distributed data scalable adaptive map-reduce framework | |
| CN107526645B (en) | A kind of communication optimization method and system | |
| JP2017529607A (en) | Compile a graph-based program specification with automated clustering of graph components based on identifying specific data port connections | |
| Belgacem et al. | A hybrid HPC/cloud distributed infrastructure: Coupling EC2 cloud resources with HPC clusters to run large tightly coupled multiscale applications | |
| Cheong et al. | SCARL: Attentive reinforcement learning-based scheduling in a multi-resource heterogeneous cluster | |
| CN109254836A (en) | Time limit committed cost Optimization Scheduling towards the preferential dependence task of cloud computing system | |
| WO2021159929A1 (en) | Topology diagram conversion system and method | |
| WO2021159928A1 (en) | Distributed signature decision-making system and method for logical nodes | |
| CN110162297A (en) | A kind of source code fragment natural language description automatic generation method and system | |
| CN110781180A (en) | Data screening method and data screening device | |
| Wang et al. | CEFS: Compute-efficient flow scheduling for iterative synchronous applications | |
| CN109976873B (en) | Scheduling scheme obtaining method and scheduling method of containerized distributed computing framework | |
| WO2022062648A1 (en) | Automatic driving simulation task scheduling method and apparatus, device, and readable medium | |
| Duan et al. | Mercury: A simple transport layer scheduler to accelerate distributed DNN training | |
| CN113010296A (en) | Task analysis and resource allocation method and system based on formalized model | |
| CN104360962A (en) | Multilevel nested data transmission method and system matched with high-performance computer structure | |
| CN111160535A (en) | DGCNN model acceleration method based on Hadoop | |
| Bestavros et al. | A Domain-Specific Language for Incremental and Modular Design of Large-Scale Verifiably-Safe Flow Networks (Preliminary Report) | |
| CN103606042B (en) | Services Composition instance migration availability deciding method based on dynamic dependency graph |
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 |