CN104184674A - Network simulation task load balancing method in heterogeneous computing environment - Google Patents

Network simulation task load balancing method in heterogeneous computing environment Download PDF

Info

Publication number
CN104184674A
CN104184674A CN201410407933.2A CN201410407933A CN104184674A CN 104184674 A CN104184674 A CN 104184674A CN 201410407933 A CN201410407933 A CN 201410407933A CN 104184674 A CN104184674 A CN 104184674A
Authority
CN
China
Prior art keywords
node
simulation
load balance
computing environment
topological
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.)
Granted
Application number
CN201410407933.2A
Other languages
Chinese (zh)
Other versions
CN104184674B (en
Inventor
王晓锋
卞娜云
刘渊
陈世云
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Jiangnan University
Original Assignee
Jiangnan University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Jiangnan University filed Critical Jiangnan University
Priority to CN201410407933.2A priority Critical patent/CN104184674B/en
Publication of CN104184674A publication Critical patent/CN104184674A/en
Application granted granted Critical
Publication of CN104184674B publication Critical patent/CN104184674B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Landscapes

  • Data Exchanges In Wide-Area Networks (AREA)

Abstract

The invention discloses a network simulation task load balancing method in the heterogeneous computing environment. The method includes the steps that 1 a heterogeneous computing environment parameter is read in; 2 an actual topological graph of network simulation is read in; 3 an algorithm parameter is initialized; 4 load balance is initialized; 5 a current load balance degree is calculated; 6 whether the current load balance reaches a threshold value or not and whether the frequency of progression-free migration reaches three or not are judged, and a result is output; 7 a computing node with the minimum routing simulation run time and a computing node with the maximum routing simulation run time are obtained; 8 a topological node to be moved is selected; 9 the topological node is moved to the computing node with the minimum routing simulation run time from the computing node with the maximum routing simulation run time; 10 whether a new result obtained after movement is optimized or not is judged, if yes, the current result is updated, and the step 6 is conducted, and otherwise, the frequency of the progression-free migration is increased by one, and the step 6 is conducted. According to the network simulation task load balancing method, computation overhead of network simulation in the heterogeneous computing environment can be effectively reduced, and meanwhile expandability can be conducted on large-scale network simulation topology and the heterogeneous computing environment.

Description

Network analog task load balance method under a kind of heterogeneous computing environment
Technical field
The present invention relates to the network analog task load balance method under a kind of heterogeneous computing environment, belong to technical field of the computer network.
Background technology
In network analog, in especially large-scale network analog, limited computational resource far can not meet huge computation requirement, so conventionally adopt the network simulator of parallel distributed.Parallel network simulation is the process that multimachine is simulated simultaneously, pass through the simulation task division of a large scale network, then each computing node calculating in cluster is responsible for simulating a part of this network, thereby expands scale and the performance of network analog.Yet for distributed network simulation, key issue is how for the effective balance simulation task of computing environment, because computing node load balance degree is higher, simulate needed running time just fewer.
In network analog running, Router Simulation and package forward simulation are two tasks quite consuming time.Router Simulation is mainly topological diagram Network Based, and in computing network simulation process, package forward is simulated required routing table information storage, and when network topology is larger, Router Simulation needs very large computing cost; Package forward dry run accounts for the significant proportion of computing cost equally, and it is mainly the behavior of each package forward in network of portraying based on discrete event simulation technology, and then deduces out the behavior of whole network.Explanation thus, promotes the load balancing degrees that Router Simulation and package forward are simulated, and further promotes simulated performance, is the key that reduces the computing cost of whole network analog.
In prior art, for balancing method of loads the present invention under homogeneous environment, claim that these class methods are uniform load balance method (ULB, Uniform Load Balance).ULB method is simulated task uniform distribution to each computing node by package forward, yet the method can effectively be divided the network analog task under homogeneous computing environment, the load balance, communication overhead and the synchronization overhead that realize simulation task minimize etc., thereby reduce the running time of network analog, but said method cannot meet the requirement of the network analog load balancing under heterogeneous computing environment.
In prior art, for balancing method of loads the present invention under isomerous environment, claim that these class methods are linear load balance method (LLB, Liner Load Balance).LLB method is according to the different computing capability of different computing nodes under isomerous environment, pro-rata network analog task, however the method can only, for guaranteeing the load balance of package forward simulation task, can not guarantee Router Simulation task load balance.
Summary of the invention
The object of the invention is to for above-mentioned the deficiencies in the prior art, a kind of heterogeneous computing environment lower network simulation task load equalization methods proposing, consider the load balance of Router Simulation task and package forward simulation task, to reduce network analog time overhead, improve Parallel Simulation efficiency.Meanwhile, the running time of the method is lower, has towards the extensibility of large scale network and large-scale calculations environment.
According to technical scheme provided by the invention, the network analog task load balance method under described heterogeneous computing environment comprises the following steps:
(1) read in a heterogeneous computing environment parameter, comprise computing node number M in heterogeneous computing environment, i computing node package forward analog capability parameter K i, i computing node Router Simulation ability function F i(n), 1≤i≤M;
(2) read in the practical topology figure G (V, E) of a network analog, its Point Set closes V={v 1, v j..., v n, some weight w v(v j), limit weight w e(v i, v j); v jrepresent j topological node, the number that N is topological node, 1≤j≤N;
(3) initiation parameter: initialization gets nowhere and moves number of times Counter is 0, and the threshold value B_Delta of initialization network analog task load equilibrium degree B is 0.95, the degre e of load balancing B of the current optimum of initialization estbe 0;
(4) load balance initialization: be { G by topological diagram G (V, E) initial division 1..., G i..., G m, make package forward simulation task load equilibrium degree BP maximum;
(5) calculate the degre e of load balancing B of current optimum est;
(6) judgement B estwhether be less than B_Delta and Counter and whether be less than 3, satisfied termination circulates, output loading balance result { G 1..., G i..., G m; Do not meet and forward step (7) to;
(7) add up current each G itopological node number n i, and according to the division Gi of Router Simulation computing cost function acquisition Router Simulation minimum running time of each computing node minwith maximum division Gi max;
(8) from the topological node v that middle selection is a certain to be moved m, v wherein mmust be with in certain node between exist link and some weights be all with exist in the node of link minimum;
(9) by node v mfrom move to and obtain new load balance result { G 1..., G i..., G m;
(10) to topological node v mnew load balance result after migration is assessed, if load balance degree B has optimization, accepts this new load balance result, upgrades B est, Counter resets to 0, goes to step (6) and carries out; Otherwise Counter adds 1, go to step (6) and carry out.
Described network analog task load equilibrium degree B calculates by following formula:
B=α×BR+(1-α)BP
Wherein, BP is package forward simulation task load equilibrium degree, and BR is Router Simulation task load equilibrium degree; α is weights, span is [0,1], as α >0.5, represent to stress to consider Router Simulation task load balance, as α <0.5, represent to stress to consider package forward simulation task load balance, if consider the load balance of Router Simulation task and package forward simulation task, choose α=0.5.
Described package forward simulation task load equilibrium degree BP obtains by following formula:
w wherein ibe i the assigned topological node of computing node weights and, K ibe i computing node package forward analog capability parameter.
Described Router Simulation task load equilibrium degree BR calculates by following formula:
n wherein ibe i the topological node number that computing node is assigned, F i(n) be i computing node Router Simulation energy force function.
Compared with prior art there is following advantage in the present invention:
1. the present invention has considered the load balancing of package forward simulation task and Router Simulation task, can effectively improve the load balancing degrees of heterogeneous computing environment lower network simulation task, the running time of reducing network analog, promotes simulated performance.
2. the present invention is by the limited load balancing that realizes Router Simulation task of migrating of node, and method computation complexity is lower, and proves by embodiment.
3. the embodiment of the present invention proves, the network analog task load balance method under heterogeneous computing environment is also effective for large scale scale heterogeneous computing environment and Large-Scale Network Simulation topology.
Accompanying drawing explanation
Fig. 1 is flow chart of the present invention.
Fig. 2 is overall dry run time comparison diagram.
Fig. 3 is method comparison diagram running time.
Fig. 4 (a)-(d) is the network analog task load equilibrium degree B comparison diagram in large scale experiment.Wherein,
Fig. 4 (a) is the load balancing degrees of 2000 router node topologys, and Fig. 4 (b) is the load balancing degrees of 3000 router node topologys, and Fig. 4 (c) is the load balancing degrees of 4000 router node topologys,
Fig. 4 (d) is the load balancing degrees of 5000 router node topologys.
Fig. 5 (a)-(d) is balancing method of loads comparison running time in large scale experiment.Wherein, Fig. 5 (a) is the method running time of 2000 router node topologys, Fig. 5 (b) is the method running time of 3000 router node topologys, Fig. 5 (c) is the method running time of 4000 router node topologys, and Fig. 5 (d) is the method running time of 5000 router node topologys.
Embodiment
Below in conjunction with drawings and Examples, the present invention is described in further detail.
The present invention can be used for the network analog task load balance under heterogeneous computing environment, promotes performance of network simulation, as shown in Figure 1, comprises the following steps:
Step 1, reads in a heterogeneous computing environment parameter, comprises computing node number: M in heterogeneous computing environment; I computing node package forward analog capability: K i(1≤i≤M); I computing node Router Simulation energy force function: F i(n) (1≤i≤M).
In the embodiment of the present invention, 4 computing nodes in the heterogeneous computing environment adopting, its basic configuration is as follows: the cpu type of computing node 1 is 1 Intel Pentium (R), 4 processors, inside saves as 2GB; The cpu type of computing node 2 is 1 AMD6128 processor, inside saves as 4GB; The cpu type of computing node 3 is 2 Intel E5520 processors, inside saves as 8GB; The cpu type of computing node 4 is 2 Intel E5620 processors, inside saves as 16GB; On 4 computing nodes, all adopt the distributed network analog platform dry run based on PDNS.Therefore, the heterogeneous computing environment computing node number M=4 adopting, its package forward analog capability is respectively K 1=505,330 package forward number/seconds, K 2=841,300 package forward number/seconds, K 3=1,248,500 package forward number/seconds, K 4=1,367,400 package forward number/seconds.The Router Simulation energy force function of 4 computing nodes is respectively:
F 1 ( n ) = 68.9 n 2 + 553.9 n + 574.9 0 &le; n &le; 3300 31.7 n 2 + 737.9 n + 6887.3 n > 3300
F 2 ( n ) = 55.6 n 2 + 600 . 7 n + 637.3 0 &le; n &le; 3300 82 . 5 n 2 + 767.1 n + 8318 . 1 n > 3300
F 1 ( n ) = 68.9 n 2 + 553.9 n + 574.9 0 &le; n &le; 3300 31.7 n 2 + 737.9 n + 6887.3 n > 3300
F 4 ( n ) = 66 . 2 n 2 + 370.5 n + 353.8 0 &le; n &le; 3300 23 . 3 n 2 + 427.1 n + 4637.0 n > 3300
Step 2, reads in the practical topology figure G (V, E) of a network analog, and its Point Set closes: V={v 1, v j..., v n, some weights: w v(v j), limit weights: w e(v i, v j).
In embodiments of the present invention, by NEM topology Core Generator, generate 5 groups of practical topology figure, there are respectively 300,600,900,1200,1500 router nodes, and the node that in above topology, each degree is 1 is connected to 28 host nodes, the 5 group network topologys that form are as follows: group 1 (Set 1) has 4,472 nodes; Group 2 (Set 2) have 9,028 nodes; Group 3 (Set 3) have 13,528 nodes; Group 4 (Set 4) have 17,804 nodes; Group 5 (Set 5) have 22,108 nodes.For above-mentioned each group network topology, be set as follows network analog task: each main frame selects another main frame to carry out FTP transfer of data at random, it is 100 seconds that the transmitting continuous time is set.The point weights of above-mentioned each group network topology are used for portraying network analog topology topological node v jpackage forward simulation task amount, these weights are as the basis of package forward simulation task load balance, limit weights are w e(v i, v j) for portraying network analog topology topological node v iwith v jbetween the package forward amount of link, these weights are as the minimized basis of communication overhead.
Step 3, initiation parameter.
Initialization gets nowhere and moves number of times Counter is 0, and the threshold value B_Delta of initialization network analog task load equilibrium degree B is 0.95, the current optimal load degree of balance of initialization B estbe 0.
Step 4, load balance initialization: using METIS dividing tool, is { G by topological diagram G (V, E) initial division 1, G 2, G 3, G 4, make package forward simulation task load equilibrium degree BP maximum.
Step 5, calculates the degre e of load balancing B of current optimum est.
B=α×BR+(1-α)BP
Wherein α is weights, span is [0,1], as α >0.5, represent to stress to consider Router Simulation task load balance, as α <0.5, represent to stress to consider package forward simulation task load balance, if consider the load balance of Router Simulation task and package forward simulation task, choose α=0.5.
In the embodiment of the present invention, optimum degre e of load balancing B estcalculate as follows:
B est=α * BR+ (1-α) BP, wherein α gets 0.5.
Wherein BR = min { F 1 ( n 1 ) , F 2 ( n 2 ) , F 3 ( n 3 ) , F 4 ( n 4 ) } max { F 1 ( n 1 ) , F 2 ( n 2 ) , F 3 ( n 3 ) , F 4 ( n 4 ) } , BP = min { w 1 / K 1 , w 2 / K 2 , w 3 / K 3 , w 4 / K 4 } max { w 1 / K 1 , w 2 / K 2 , w 3 / K 3 , w 4 / K 4 } .
W ibe i the assigned topological node of computing node weights and, K ibe i computing node package forward analog capability parameter, n ibe i the topological node number that computing node is assigned, F i(n) be i computing node Router Simulation energy force function.
Step 6, judgement B estwhether be less than B_Delta (0.95) and Counter and whether be less than 3, satisfied termination circulation is exported and is divided (load balance result) { G 1, G 2, G 3, G 4; Do not meet and forward step (7) to.
Step 7, adds up each G itopological node number n i(1≤i≤4), and each computing node Router Simulation energy force function reading according to step 1 obtains the division of Router Simulation minimum running time with maximum division
When topological node moves, first judge the division of Router Simulation computing cost minimum with maximum division to determine the migratory direction of topological node, should be from middle selection node motion arrives
In embodiments of the present invention, by following formula, obtain: 1≤i min≤ 4 and satisfied: F i min ( n i min ) = min { F i ( n i ) | 1 &le; i &le; 4 } , 1 &le; i max &le; 4 And meet: F i max ( n i max ) = max { F i ( n i ) | 1 &le; i &le; 4 } .
Step 8, from the topological node v that middle selection is a certain to be moved m, v wherein m(v mmay exist a plurality ofly, choose at random one of them) must be with in certain node between exist link and some weights be all with exist in the node of link minimum.
In embodiments of the present invention, choose by the following method:
First choose both candidate nodes collection: and make (v i, u j) ∈ G};
Then based on node weights screening both candidate nodes collection:
V' c={v i|(v i∈V c)∧(w v(v i)=min{w v(v j)v j∈V c})};
Last random selected v m, v wherein m∈ V ' c, and meet:
As described above, topological node v mconcrete choosing method as follows: first choose and belong to topological node collection V c, the node in this set must be with in certain node between there is link; Then for V c, choose the topological node of its mid point weights minimum, and form set V ' c(topological node of some weights minimum may have a plurality of), why the topological node of selected point weights minimum, is because will make topological node migration minimum on the impact of package forward simulation task load balance; Finally choose V ' cin with in between other topological nodes link metric and minimum topological node (if there are a plurality of nodes that satisfy condition, choosing at random one of them) as the node v of most suitable movement m, why so selecting is because will make the rear newly-increased communication overhead of topological node migration minimum.
Step 9, by node v mfrom move to and obtain new load balance result { G 1, G 2, G 3, G 4;
Step 10, to topological node v mnew load balance result after migration is assessed, if load balance degree B has optimization, accepts this new load balance result, upgrades B est, Counter resets to 0, goes to step 6 execution.Otherwise Counter adds 1, go to step 6 execution.
Effect of the present invention can further illustrate by following l-G simulation test.
Embodiment mono-
Existing parallel network simulation software, comprises PDNS, and GTNETS and Genesis etc., in example of the present invention, adopt PDNS to verify as parallel network simulation device.The load balance result that said method is obtained, runtime verification in the distributed network analog platform based on PDNS.
The present invention is based on above-mentioned experimental situation, by many groups, test to verify that method proposed by the invention (using LBRP reduced representation) is with respect to the validity of ULB and LLB method.Adopt respectively ULB, LLB and LBRP method to divide for above-mentioned 5 group network simulation tasks, and the simulation of the experimental situation operational network based on above-mentioned task, record the running time that overall dry run time and load balance result obtain: Fig. 2 is total dry run time comparison diagram; Fig. 3 is comparison diagram running time that load balance result obtains.
Fig. 2 has shown that LBRP can effectively reduce the universe network dry run time: with respect to ULB, 5 groups of empirical averages have reduced by 31.6%; With respect to LLB, 5 groups of empirical averages have reduced by 16.8%.This main cause is the load balance that this method has been taken into account Router Simulation and package forward simulation, therefore can effectively reduce the universe network dry run time.
Fig. 3 has compared comparison diagram running time of the load balance result acquisition of 3 kinds of methods such as ULB, LLB and LBRP, as seen from the figure, with respect to ULB, LLB, LBRP needed time when carrying out load balance obviously increases, and this is mainly because LBRP need to consider the package forward simulation load balancing of task and the load balancing of Router Simulation task simultaneously.From Fig. 3, divide the total amount increasing running time, LBRP is no more than 0.1 second needed extra time, and from Fig. 2, the simulation Runtime that LBRP reduces can reach even several kiloseconds of hundreds of.Therefore, LBRP, in a small amount of increase of needed running time in task division stage, exchanges a large amount of minimizings of simulation Runtime for, is worth very much.
Embodiment bis-
Large scale scale heterogeneous computing environment in this example obtains by the virtual extended of heterogeneous computing environment in embodiment mono-, always have 5 groups of heterogeneous computing environments, its scale is respectively 10,20,30,40,50 computing nodes, and the concrete computing node performance that each scale comprises is as shown in table 1.The computing environment scale 10 in table 1 of take is example, show that this computing environment has 10 computing nodes, wherein in the performance of 2 computing nodes (comprising package forward analog capability and Router Simulation computing cost) and embodiment mono-, computing node 1 is consistent, 3 consistent with computing node 2,3 consistent with computing node 3, and 2 consistent with computing node 4.
The structure of the large scale scale heterogeneous computing environment of table 1
Computing environment scale Computing node 1 Computing node 2 Computing node 3 Computing node 4
10 2 3 3 2
20 5 5 5 5
30 7 8 8 7
40 10 10 10 10
50 12 13 13 12
Aspect network analog topology, by NEM topology Core Generator, generate 4 groups of topologys, there are respectively 2000,3000,4000,5000 router nodes, and the node that in above topology, each degree is 1 is connected to 28 host nodes, the 4 group network topologys that form are as follows: group 1 (Set 1) has 29,300 nodes; Group 2 (Set 2) have 44,020 nodes; Group 3 (Set 3) have 59,076 nodes; Group 4 (Set 4) have 77,800 nodes.Use respectively ULB, LLB and LBRP to carry out load balance according to computing environment given in table 1 these 4 groups of topologys.
For division result, computing network simulation task load equilibrium degree (B), comprehensively to weigh the load balance situation of Router Simulation task and package forward simulation task, comparative result is as shown in Figure 4.As can be seen from Figure 4, LBRP obviously improves with respect to the load balancing degrees of ULB and LLB: the average load equilibrium degree that the average load equilibrium degree of ULB is 19.11%, LLB is 47.20%, and the average load equilibrium degree of LBRP is 90.18%.
The running time of 3 kinds of balancing method of loads such as ULB, LLB and LBRP as shown in Figure 5.Obviously, with respect to ULB and LLB, the needed time division of LBRP obviously improves.Along with the increase of network topology scale and computing environment scale, LLB and LBRP divide also to be increased needed running time gradually, but sharply steep situation about increasing does not appear in LBRP.From the increment of running time, LBRP is no more than at most 0.4 second needed extra time, and this needs the dry run time of a few hours to compare with Large-Scale Network Simulation is inappreciable.
Complex chart 4 is visible with Fig. 5, the LBRP method proposing herein can be applicable to large-scale heterogeneous computing environment and Large-Scale Network Simulation topology: when obtaining great load balancing degrees lifting, extra division running time less (in 0.4 second) of required introducing, therefore has extensibility.

Claims (4)

1. the network analog task load balance method under heterogeneous computing environment, is characterized in that, comprises the following steps:
(1) read in a heterogeneous computing environment parameter, comprise computing node number M in heterogeneous computing environment, i computing node package forward analog capability parameter K i, i computing node Router Simulation ability function F i(n), 1≤i≤M;
(2) read in the practical topology figure G (V, E) of a network analog, its Point Set closes V={v 1, v j..., v n, some weight w v(v j), limit weight w e(v i, v j); v jrepresent j topological node, the number that N is topological node, 1≤j≤N;
(3) initiation parameter: initialization gets nowhere and moves number of times Counter is 0, and the threshold value B_Delta of initialization network analog task load equilibrium degree B is 0.95, the degre e of load balancing B of the current optimum of initialization estbe 0;
(4) load balance initialization: be { G by topological diagram G (V, E) initial division 1..., G i..., G m, make package forward simulation task load equilibrium degree BP maximum;
(5) calculate the degre e of load balancing B of current optimum est;
(6) judgement B estwhether be less than B_Delta and Counter and whether be less than 3, satisfied termination circulates, output loading balance result { G 1..., G i..., G m; Do not meet and forward step (7) to;
(7) add up current each G itopological node number n i, and according to the division of Router Simulation computing cost function acquisition Router Simulation minimum running time of each computing node with maximum division
(8) from the topological node v that middle selection is a certain to be moved m, v wherein mmust be with in certain node between exist link and some weights be all with exist in the node of link minimum;
(9) by node v mfrom move to and obtain new load balance result { G 1..., G i..., G m;
(10) to topological node v mnew load balance result after migration is assessed, if load balance degree B has optimization, accepts this new load balance result, upgrades B est, Counter resets to 0, goes to step (6) and carries out; Otherwise Counter adds 1, go to step (6) and carry out.
2. the network analog task load balance method under heterogeneous computing environment according to claim 1, its spy
Levy is that the described network analog task load equilibrium degree B of step (3) calculates by following formula:
B=α×BR+(1-α)BP
Wherein, BP is package forward simulation task load equilibrium degree, and BR is Router Simulation task load equilibrium degree; α is weights, span is [0,1], as α >0.5, represent to stress to consider Router Simulation task load balance, as α <0.5, represent to stress to consider package forward simulation task load balance, if consider the load balance of Router Simulation task and package forward simulation task, choose α=0.5.
3. according to claim 1, the network analog task load balance method under the heterogeneous computing environment described in 2, is characterized in that, described package forward simulation task load equilibrium degree BP obtains by following formula:
w wherein ibe i the assigned topological node of computing node weights and, K ibe i computing node package forward analog capability parameter.
4. the network analog task load balance method under heterogeneous computing environment according to claim 2, is characterized in that, described Router Simulation task load equilibrium degree BR calculates by following formula:
n wherein ibe i the topological node number that computing node is assigned, F i(n) be i computing node Router Simulation energy force function.
CN201410407933.2A 2014-08-18 2014-08-18 A kind of network analog task load balance method under heterogeneous computing environment Active CN104184674B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201410407933.2A CN104184674B (en) 2014-08-18 2014-08-18 A kind of network analog task load balance method under heterogeneous computing environment

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201410407933.2A CN104184674B (en) 2014-08-18 2014-08-18 A kind of network analog task load balance method under heterogeneous computing environment

Publications (2)

Publication Number Publication Date
CN104184674A true CN104184674A (en) 2014-12-03
CN104184674B CN104184674B (en) 2017-04-05

Family

ID=51965432

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201410407933.2A Active CN104184674B (en) 2014-08-18 2014-08-18 A kind of network analog task load balance method under heterogeneous computing environment

Country Status (1)

Country Link
CN (1) CN104184674B (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105515818A (en) * 2015-06-11 2016-04-20 哈尔滨安天科技股份有限公司 Method and system for splitting cyclic structure in network topology
CN106888115A (en) * 2017-02-09 2017-06-23 中国科学院信息工程研究所 A kind of constructing network topology method and system

Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1687917A (en) * 2005-05-11 2005-10-26 上海理工大学 Large scale data parallel computing main system and method under network environment

Patent Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1687917A (en) * 2005-05-11 2005-10-26 上海理工大学 Large scale data parallel computing main system and method under network environment

Non-Patent Citations (4)

* Cited by examiner, † Cited by third party
Title
朱伟 等: ""异构计算环境下网络路由模拟任务的非线性划分"", 《系统仿真学报》 *
王晓锋 等: ""基于快速网络模拟的安全态势预测"", 《系统仿真学报》 *
王晓锋 等: ""基于拓扑抽象的高性能网络模拟方法"", 《计算机工程与应用》 *
王晓锋 等: ""核心节点全局计算与存储的路由模拟策略"", 《计算机工程与应用》 *

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105515818A (en) * 2015-06-11 2016-04-20 哈尔滨安天科技股份有限公司 Method and system for splitting cyclic structure in network topology
CN105515818B (en) * 2015-06-11 2019-07-02 哈尔滨安天科技股份有限公司 The method and system of cyclic structure are split in a kind of network topology layout
CN106888115A (en) * 2017-02-09 2017-06-23 中国科学院信息工程研究所 A kind of constructing network topology method and system
CN106888115B (en) * 2017-02-09 2019-08-02 中国科学院信息工程研究所 A kind of constructing network topology method and system

Also Published As

Publication number Publication date
CN104184674B (en) 2017-04-05

Similar Documents

Publication Publication Date Title
CN108009016B (en) Resource load balancing control method and cluster scheduler
CN102866915B (en) Virtual cluster integration method, device and system of virtual cluster
CN105550323A (en) Load balancing prediction method of distributed database, and predictive analyzer
CN103347055B (en) Task processing system in cloud computing platform, Apparatus and method for
Deng et al. Heat diffusion based dynamic load balancing for distributed virtual environments
CN104216784A (en) Hotspot balance control method and related device
US11055139B2 (en) Smart accelerator allocation and reclamation for deep learning jobs in a computing cluster
CN107609141A (en) It is a kind of that quick modelling method of probabilistic is carried out to extensive renewable energy source data
CN110415160B (en) GPU (graphics processing Unit) topology partitioning method and device
Sahman et al. The binary differential search algorithm approach for solving uncapacitated facility location problems
CN107329826A (en) A kind of heuristic fusion resource dynamic dispatching algorithm based on Cloudsim platforms
CN104184674A (en) Network simulation task load balancing method in heterogeneous computing environment
Jiang et al. Hierarchical deployment of deep neural networks based on fog computing inferred acceleration model
CN104778088A (en) Method and system for optimizing parallel I/O (input/output) by reducing inter-progress communication expense
Huang et al. An efficient parallel method for batched OS-ELM training using MapReduce
CN116303219A (en) Grid file acquisition method and device and electronic equipment
Samad et al. Two round scheduling (TRS) scheme for linearly extensible multiprocessor systems
CN115001978B (en) Cloud tenant virtual network intelligent mapping method based on reinforcement learning model
Kaveh et al. Bandwidth, profile and wavefront optimization using PSO, CBO, ECBO and TWO algorithms
Jansson High performance adaptive finite element methods for turbulent fluid flow
Biswas et al. Global load balancing with parallel mesh adaption on distributed-memory systems
Djomehri et al. Load balancing strategies for multi-block overset grid applications
Czekster et al. Kronecker descriptor partitioning for parallel algorithms
CN117435308B (en) Modelica model simulation method and system based on parallel computing algorithm
Zhu et al. A Hopfield neural network based task mapping method

Legal Events

Date Code Title Description
C06 Publication
PB01 Publication
C10 Entry into substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant