+ All Categories
Home > Documents > Grid Load Balancing Scheduling Algorithm Based on Statistics Thinking The 9th International...

Grid Load Balancing Scheduling Algorithm Based on Statistics Thinking The 9th International...

Date post: 21-Dec-2015
Category:
View: 217 times
Download: 2 times
Share this document with a friend
21
Grid Load Balancing Scheduling Algorithm Based on Statistics Thinking The 9th International Conference for Young Computer Scientists Bin Lu, Hongbin Zhang Present by Chen, Ting-Wei
Transcript
Page 1: Grid Load Balancing Scheduling Algorithm Based on Statistics Thinking The 9th International Conference for Young Computer Scientists Bin Lu, Hongbin Zhang.

Grid Load Balancing Scheduling Algorithm Based on Statistics

ThinkingThe 9th International Conference for Young Computer Scientists

Bin Lu, Hongbin Zhang

Present by Chen, Ting-Wei

Page 2: Grid Load Balancing Scheduling Algorithm Based on Statistics Thinking The 9th International Conference for Young Computer Scientists Bin Lu, Hongbin Zhang.

2

Table of content

• Introduction

• Hierarchical Grid system model

• Grid resource allocation and task scheduling

• Simulative experiments

• Conclusions

Page 3: Grid Load Balancing Scheduling Algorithm Based on Statistics Thinking The 9th International Conference for Young Computer Scientists Bin Lu, Hongbin Zhang.

3

Introduction

• Propose a new task scheduling and resource allocation algorithm– Load tracking module– Job distributing module– Load monitoring module

• Simulate the algorithm

Page 4: Grid Load Balancing Scheduling Algorithm Based on Statistics Thinking The 9th International Conference for Young Computer Scientists Bin Lu, Hongbin Zhang.

4

Introduction (cont.)

• Based on the hierarchical Grid structure model– Distribution of Grid resources – The strategy of task scheduling– Following statistical thinking

Page 5: Grid Load Balancing Scheduling Algorithm Based on Statistics Thinking The 9th International Conference for Young Computer Scientists Bin Lu, Hongbin Zhang.

5

Hierarchical Grid system model

Page 6: Grid Load Balancing Scheduling Algorithm Based on Statistics Thinking The 9th International Conference for Young Computer Scientists Bin Lu, Hongbin Zhang.

6

Grid resource allocation and task scheduling

Load (LLoad (Lxx))Length of CPU Length of CPU Ready queueReady queue

CPU UtilizationCPU Utilization

Light LoadLight Load LLxx < Llow Short Low

Heavy LoadHeavy Load LLxx >Lhigh Long High

Moderate Moderate LoadLoad

Llow < LLxx < Lhigh Normal Normal

• Definition of the Load

Page 7: Grid Load Balancing Scheduling Algorithm Based on Statistics Thinking The 9th International Conference for Young Computer Scientists Bin Lu, Hongbin Zhang.

7

Grid resource allocation and task scheduling (cont.)

• Algorithm analysis and implementation– Three components

• Load tracing module

• Job distributing module

• Load monitor module

Page 8: Grid Load Balancing Scheduling Algorithm Based on Statistics Thinking The 9th International Conference for Young Computer Scientists Bin Lu, Hongbin Zhang.

8

Grid resource allocation and task scheduling (cont.)

– Three tables are installed on the database server in GRM and DRM

• The first: Every sub-DRM load information

• The second: The number of tasks being operated

• The third: The throughput and the average delayed response

Page 9: Grid Load Balancing Scheduling Algorithm Based on Statistics Thinking The 9th International Conference for Young Computer Scientists Bin Lu, Hongbin Zhang.

9

Grid resource allocation and task scheduling (cont.)

• Load tracing module– Get the load information of DRM– Monitor periodically the implementation state o

f operations– Report it to the upper level for the overall mana

gement and scheduling

Page 10: Grid Load Balancing Scheduling Algorithm Based on Statistics Thinking The 9th International Conference for Young Computer Scientists Bin Lu, Hongbin Zhang.

10

Grid resource allocation and task scheduling (cont.)

• Job distributing module– According to load balancing– GRM chooses a suitable DRM to carry out job

distributing– Consider

• CPU utilization • Memory usage

– The weight of load indicators can be determined statically or dynamically

Page 11: Grid Load Balancing Scheduling Algorithm Based on Statistics Thinking The 9th International Conference for Young Computer Scientists Bin Lu, Hongbin Zhang.

11

Grid resource allocation and task scheduling (cont.)

– Load computing formula of the CN

– The average of the sub-DRM’s load or CN’s load

– Variance of load

2

1

n

i j ijj

L a L

1

1 m

DK ii

L Lm

2( ) [( ) ]DK i DKVar L E L L

Page 12: Grid Load Balancing Scheduling Algorithm Based on Statistics Thinking The 9th International Conference for Young Computer Scientists Bin Lu, Hongbin Zhang.

12

Grid resource allocation and task scheduling (cont.)

– Job algorithmInitialize;For (;;){do

If (There was information form the DRMs or CNs){

Updating the load information of DRM or CNs according to the formula (1)-(3);Calculating the rate of throughput;Calculating the average response delay;If (There was information of completed assignment){

Deducting one assignment of corresponding DRM;

}}

Page 13: Grid Load Balancing Scheduling Algorithm Based on Statistics Thinking The 9th International Conference for Young Computer Scientists Bin Lu, Hongbin Zhang.

13

Grid resource allocation and task scheduling (cont.)

If (Portal gives the assignment to GRM){

If (The node having the load is not unique){ ;Making the node whose variance is as the scheduling nodes, if still not unique, selecting the first node;}Else{Making the node whose load value is be the scheduling node;}Giving it the assignment for scheduling, and updating the number of assignment of DRM;}

}

min min kk

L L

minmax max

DKDK

L LVar Var

maxVar

minL

minL

Page 14: Grid Load Balancing Scheduling Algorithm Based on Statistics Thinking The 9th International Conference for Young Computer Scientists Bin Lu, Hongbin Zhang.

14

Grid resource allocation and task scheduling (cont.)

• Load monitor module– Surveillance

• The current load of all the DRMs

Page 15: Grid Load Balancing Scheduling Algorithm Based on Statistics Thinking The 9th International Conference for Young Computer Scientists Bin Lu, Hongbin Zhang.

15

Grid resource allocation and task scheduling (cont.)

For (;;){

If (sum < ){

The opening of stand-by resources manager;}Reading the load vector of domain resources manager;Reading the upper and lower threshold provided in profile;If (Every load exceeds the threshold ){

Opening the stand-by resources manager;}If (Every load is below the threshold ){

Removing stand-by resources manager;}

mb

highL

lowL

Page 16: Grid Load Balancing Scheduling Algorithm Based on Statistics Thinking The 9th International Conference for Young Computer Scientists Bin Lu, Hongbin Zhang.

16

Grid resource allocation and task scheduling (cont.)

Else{

Get the last updated time of the nodeIf (The current time – last updated time) > (timeout)Means that the main resource manager failed and it takes proper measures;

}}

Page 17: Grid Load Balancing Scheduling Algorithm Based on Statistics Thinking The 9th International Conference for Young Computer Scientists Bin Lu, Hongbin Zhang.

17

Simulative experiments

• Use Gridsim to simulate

• Compare performance of the load balancing scheduling method– Min-Min algorithm– Ant algorithm– This algorithm (This paper proposed)

Page 18: Grid Load Balancing Scheduling Algorithm Based on Statistics Thinking The 9th International Conference for Young Computer Scientists Bin Lu, Hongbin Zhang.

18

Simulative experiments (cont.)

• Configuration of experiment environment

GRM

DRM1 DRM2 DRM3

DRM11 DRM12 DRM13 CN-17台CN-21台

CN-18台 CN-25台 CN-28台

USER

Page 19: Grid Load Balancing Scheduling Algorithm Based on Statistics Thinking The 9th International Conference for Young Computer Scientists Bin Lu, Hongbin Zhang.

19

Simulative experiments (cont.)

• Comparison of experiment results

Page 20: Grid Load Balancing Scheduling Algorithm Based on Statistics Thinking The 9th International Conference for Young Computer Scientists Bin Lu, Hongbin Zhang.

20

Conclusions

• Statistical load balancing scheduling scheme

• Future work– Granular computing idea to improve this

algorithm– More the flexibility and efficiency of this

algorithm

Page 21: Grid Load Balancing Scheduling Algorithm Based on Statistics Thinking The 9th International Conference for Young Computer Scientists Bin Lu, Hongbin Zhang.

Thank you for your attentionThank you for your attention


Recommended