Joon Bum Kim, Su Kyung Hwang, Sung Ho Jung, Suk Jung Choo , Cheol Hyun Chung, Jae Won Lee
A Distributed Load Balancing Approach for Industrial IEEE 802.11 Wireless Networks 2014 YU-ANTL...
-
Upload
martin-norman -
Category
Documents
-
view
217 -
download
0
Transcript of A Distributed Load Balancing Approach for Industrial IEEE 802.11 Wireless Networks 2014 YU-ANTL...
A Distributed Load Balancing Approachfor Industrial IEEE 802.11 Wireless
Networks
A Distributed Load Balancing Approachfor Industrial IEEE 802.11 Wireless
Networks
2014 YU-ANTL Seminal
Hyun dong HwangAdvanced Networking Technology Lab. (YU-ANTL)
Dept. of Information & Comm. Eng, Graduate School, Yeungnam University, KOREA
(Tel : +82-53-810-3940; Fax : +82-53-810-4742http://antl.yu.ac.kr/; E-mail : [email protected])
Advanced Networking Tech. Lab.Yeungnam University (YU-ANTL)
YU-ANTL Lab SeminalHyun dong Hwang2
Outline
Introduction System Model Dynamic Load Balancing Algorithm
Performance Evaluation First scenario Second scenario
Conclusions and Future Works
Advanced Networking Tech. Lab.Yeungnam University (YU-ANTL)
YU-ANTL Lab SeminalHyun dong Hwang3
Introduction
Introduction Load balancing is a technique used to balance traffic flows in IEEE
wireless networks in order to increase throughput. When network load exceeds, or is close to its maximum capacity,
congestion occurs and throughput decreases. load balancing technique comes into play when different APs
cover the same area or there is an overlapping area where hosts can choose to connect to, at least, two APs.
Load balancing approaches classification Centralized approach
Pros : Ensures devices interoperability. Cons : Entire system failure, while network size increases, even central
node load processing increases at the expense of its efficiency. Distributed approach
Pros : Fault-tolerant, not require the use of additional network entities for APs management
Cons : coordination need among APs in order to ensure information in-tegrity, accuracy and consistency about network status.
Advanced Networking Tech. Lab.Yeungnam University (YU-ANTL)
YU-ANTL Lab SeminalHyun dong Hwang4
Introduction
Selection and transfer techniques : About the selection of the new station, two approaches can be used Random selection : AP is randomly chosen as candidate for
transfer Pros : Technique is simple to implement and does not require high
computational time Cons : It may not be the best for load balance achieving in the short-
est possible time. Best candidate : the most appropriate station is selected taking
into account load metrics like traffic information generated by the station itself, packet loss and the average network traffic. Threshold-based : AP can accept a station association request if an
established load metric does not exceed a threshold value decided a priori
Relative threshold : load metric is evaluated considering other APs load and then the best one is chosen
Advanced Networking Tech. Lab.Yeungnam University (YU-ANTL)
YU-ANTL Lab SeminalHyun dong Hwang5
Introduction
Load displacement mechanisms There are three basic approaches for load control
Association management– Overloaded AP can send a dissociation frame to an already associated sta-
tion, hoping for a re-association to another AP more appropriate Admission control– AP can simply refuse new association requests if there is overload risk. The
request can be accepted only if predicted load is less than a threshold value Coverage adjustment– Overloaded APs can reduce transmission power of their beacon frame in or-
der to be hardly detected by new stations
Advanced Networking Tech. Lab.Yeungnam University (YU-ANTL)
YU-ANTL Lab SeminalHyun dong Hwang6
System Model
Current shortcoming In industrial networks, packets transmission is strictly related to
real-time constraints. Existing load balancing approaches, de-scribed in the previous section, are based on load metrics like throughput, packets loss or number of connected stations.
However, in industrial wireless networks (soft real-time), system efficiency can’t be measured only by throughput or packets loss percentage.
It’s necessary to evaluate QoS performances in terms of num-ber of deadline miss for each station in a given moment.
Advanced Networking Tech. Lab.Yeungnam University (YU-ANTL)
YU-ANTL Lab SeminalHyun dong Hwang7
System Model
Reference paper “Dynamic load balancing techniques for flexible wireless indus-
trial networks” Dynamic load balancing approach used in industrial real-time con-
texts. Each AP, connected to the backbone, communicates with a network
controller having a global network view and realizes corrective ac-tions in case of performances degradation.
This approach presents the same problems of centralized load balanc-ing algorithm described previously.(hardware architecture must be complex in terms of computational costs.)
Advanced Networking Tech. Lab.Yeungnam University (YU-ANTL)
YU-ANTL Lab SeminalHyun dong Hwang8
System Model
Propose system model Load balancing decisions are performed by each AP, in a dis-
tributed way, and not by a network controller only. Main approach is to provide a mechanism for load distribution
in order to obtain less deadline miss possible, lower than a tol-erable threshold.
Advanced Networking Tech. Lab.Yeungnam University (YU-ANTL)
YU-ANTL Lab SeminalHyun dong Hwang9
System Model
Propose dynamic Load balancing (Dynamic Load Bal-ancing Algorithm : DLBA) Load metric takes into account the number of Deadline Miss
(DM) measured by each station.
Advanced Networking Tech. Lab.Yeungnam University (YU-ANTL)
YU-ANTL Lab SeminalHyun dong Hwang10
Dynamic Load Balancing Algorithm
DLBA runs inside each AP in a distributed manner, chose a station from connected stations list.
Station handle condition Signal quality is lower than a
given threshold value. This occurs when a station is moving from an AP towards another one and have to perform handover to not lose connectivity.
DM exceeds a threshold value. In this case, the wireless channel is too busy, collisions cause delivery delays of soft real-time traffic flows and number of deadline miss increases.
Advanced Networking Tech. Lab.Yeungnam University (YU-ANTL)
YU-ANTL Lab SeminalHyun dong Hwang11
Dynamic Load Balancing Algorithm
Select Access Point Algorithm APs exchange each other network
information through the wired backbone, the DLBA evaluates network information of APs de-tected by managed station and signal quality that it detects.
The algorithm verifies if detected signal level is enough for connec-tion, performing a control with a threshold value.
Measures best performances, in terms of deadline miss of con-nected stations, is chosen.
Advanced Networking Tech. Lab.Yeungnam University (YU-ANTL)
YU-ANTL Lab SeminalHyun dong Hwang12
Dynamic Load Balancing Algorithm
Handover Algorithm
Advanced Networking Tech. Lab.Yeungnam University (YU-ANTL)
YU-ANTL Lab SeminalHyun dong Hwang13
Performance Evaluation
Test System Routers/APs Cisco Linksys WRT54GL(Linux system, openWRT)
First scenario 2 Linksys router in AP mode and connected through a wired
backbone Packet size is 5152 byte and transmission period was set to 10
ms, equal to relative deadline. 600 packets for second are transmitted
Advanced Networking Tech. Lab.Yeungnam University (YU-ANTL)
YU-ANTL Lab SeminalHyun dong Hwang14
Performance Evaluation
Advanced Networking Tech. Lab.Yeungnam University (YU-ANTL)
YU-ANTL Lab SeminalHyun dong Hwang15
Performance Evaluation
Throughput/Workload versus Packet Error Rate per-centage
Advanced Networking Tech. Lab.Yeungnam University (YU-ANTL)
YU-ANTL Lab SeminalHyun dong Hwang16
Performance Evaluation
DeadLine Miss measured(1)
Advanced Networking Tech. Lab.Yeungnam University (YU-ANTL)
YU-ANTL Lab SeminalHyun dong Hwang17
Performance Evaluation
DeadLine Miss measured(2)
Advanced Networking Tech. Lab.Yeungnam University (YU-ANTL)
YU-ANTL Lab SeminalHyun dong Hwang18
Performance Evaluation
Second scenario
Advanced Networking Tech. Lab.Yeungnam University (YU-ANTL)
YU-ANTL Lab SeminalHyun dong Hwang19
Performance Evaluation
Scenario 2: Throughput/Workload vs. Packet Error rate percentage
Advanced Networking Tech. Lab.Yeungnam University (YU-ANTL)
YU-ANTL Lab SeminalHyun dong Hwang20
Performance Evaluation
HOSTS MEAN DEADLINE MISS RATIO USING DLBA
Advanced Networking Tech. Lab.Yeungnam University (YU-ANTL)
YU-ANTL Lab SeminalHyun dong Hwang21
Conclusions and Future Works
The use of load metrics like workload are acceptable only when communications among nodes are not characterized by real-time constraints typical of industrial process control.
APs must always ensure best performances and react to system degradations. To this end, we have chosen Deadline Miss (DM) as load metric in a distributed approach.
Network load is equally distributed based on number of deadline miss detected by each host and on signal power of each AP. Results, obtained through measures in some real scenarios, are very promising and show significant im-provements compared to not real-time approaches and real-time centralized approaches.
Advanced Networking Tech. Lab.Yeungnam University (YU-ANTL)
YU-ANTL Lab SeminalHyun dong Hwang22
Performance Evaluation
DeadLine Miss measured(1)
Advanced Networking Tech. Lab.Yeungnam University (YU-ANTL)
YU-ANTL Lab SeminalHyun dong Hwang23
Reference
[1] IEEE Standard for Local and metropolitan area networks – part 16: Air interface for Broadband Wireless Access System, 2009.[2] A. Balachandran, P. Bahl, G. M. Voelker, “Hot-Spot Congestion Relief in Public-area Wireless Networks,” Proc. of 4th IEEE Workshop on Mobile Computing Systems and Applications, June 2002.[3] H. Velayos, V. Aleo, and G. Karlsson, “Load Balancing in Overlapping Wireless LAN Cells,” in proc. of IEEE ICC’04, pp. 3833–3836, 2004.[4] H.M. ElBadawy, “Ptimal RAT selection algorithm trhough Common radio resource management in heterogeneous wireless networks”, 28th National Radio Science Conference (NRSC), pp. 1-9, 2011[5] Xu Fengyuan, C.C. Tan, Li Qun, Yan Guanhua, Wu Jie, “Designing a Practical Access Point Association Protocol”, Proc of IEEE INFOCOM 2010, pp 1-9, 2010[6] Kuo-Shu Huang, I-Ping Hsieh, Shang-Juh Kao, “Incorporating AP selection and call admission control for seamless handoff procedure”, International Conference on Computer and Communication Engineering, pp. 823-826, 2008