Upload
jason-sims
View
219
Download
0
Tags:
Embed Size (px)
Citation preview
Dynamic Load Sharing and
BalancingHaidong Xue
Part One: Knowledge in Textbook Motivation, Models, Concepts, Algorithms
Part Two: Recent Work• Use the Mobile Agent and Include I/O• Duplex Loading Balancing Strategy• Stability Analysis Based• Extension-based Dynamic Load Balancing
Part Three: Future Work
Content
Part One: Knowledge in the Textbook
Motivations,ModelsConcepts, Algorithms
For most distributed applications: The prior knowledge of processes is not realistic
Motivation
Goal:
Reduce the overall turnaround time
= communication overhead + computational constraints
Motivation
(5.1c)
Models – Disjoint Process Model
(5.3c)
Models – Workstation Model
Load Sharing Joining the shortest queue (Randy Chow, Theodore Johnson, 1997)
Load Balancing Equalizing queue sizes (Randy Chow, Theodore Johnson, 1997)
Concepts
Sender-initiated Algorithm Receiver-initiated Algorithm
Algorithms
Transfer policy: When does a node become the sender? SQ ST
Selection policy: How does the sender choose a process for transfer? The last one.
Location policy: Which node should be the target receiver? RQ PL
Sender-initiated Algorithm
Sender-initiated Algorithm
11
24
35
44
ST=3
52
SQ = sender’s queue sizeST = sender thresholdRQ = receiver queue sizePL = poll limit
Sender-initiated Algorithm
Receiver-initiated Algorithm
16
24
35
44
ST=3
57
Receiver-initiated Algorithm
11
24
35
44
ST=3
52
Algorithms
Depends on Load Stock Exchange
Algorithms - Hybrid
Part Two: Recent Work
•Use the Mobile Agent and Include I/O•Duplex Loading Balancing Strategy•Stability Analysis Based•Extension-based Dynamic Load Balancing
In order to distribute the incoming load on participating nodes, use mobile agent (MA). A value function which is sum of all three
types of resources including CPU, Memory and I/O is considered.
(Neeraj Nehra, R.B.Patel, 2007)
Use the Mobile Agent and Include I/O
When comes to OBS, an Object-based Storage System, most of these strategies are hard to comply with the new feature
DLBS can be utilized in the real OBS system to provide more effective and efficient load balance than other methods
(Tan Zhipeng et al, 2009 )
Duplex Loading Balancing Strategy
the unstable behavior such as oscillatory action which caused by delay can intensely influence the performance of load balancing algorithm.
a linear dynamic load balancing model
(Qingyang Meng et al, 2008)
Stability Analysis Based
Connect extension theory with load Extension engineering method is initially
proposed by Prof. Cai Wen
(Zhang Xiayu et al, 2007)
Extension-based Dynamic Load Balancing
Part Three: Future Work
•Other theories•Collect more information
Incorporate other theory Collect more information of the processes Develop load balancing algorithm based on
those information
Future Work
Randy Chow, Theodore Johnson. Distributed Operating Systems& Algorithms. Addison Wesley, 1997
Neeraj Nehra, R.B. Patel. Towards Dynamic Load Balancing in Heterogeneous Cluster using Mobile Agent. International Conference on Computational Intelligence and Multimedia Applications 2007, 2007, 15-21
Tan Zhipeng, Feng Dan, Tu Xudong, He Fei. DLBS: Duplex Loading Balancing Strategy on Object Storage System. 2009 IEEE International Symposium on Parallel and Distributed Processing with Applications, 2009, 45-52
Reference
Qingyang Meng, Jianzhong Qiao, Jun Liu, Shukuan Lin. A Dynamic Load Balancing Method Based on Stability Analysis. 2008 International Symposium on Computer Science and Computational Technology, 2008, 404-408
Zhang Xiayu, Yu Yongquan, Chen Baixing, Ye Feng, Tan Xingxing. An Extension-based Dynamic Load Balancing Model of Heterogeneous Server Cluster. 2007 IEEE International Conference on Granular Computing, 2007, 675-679
Reference Contd.