Application of Fuzzy Logic in Load Balancing of Homogenous Distributed Systems1

Ali M. Alakeel
International Journal of Computer Science and Security (IJCSS), Volume (10) : Issue (3) : 2016 95
Application of Fuzzy Logic in Load Balancing of Homogenous
Distributed Systems1
Ali M. Alakeel alakeel@ut.edu.sa
Department of Computer Science
Faculty of Computing and Information Technology
University of Tabuk
P.O.Box 741, Tabuk 71491, Saudi Arabia
Abstract
Various studies have shown that distributing the work load evenly among processors of a
distributed system highly improves system performance and increases resource utilization. This
process is known as load balancing. Fuzzy logic has been applied in many fields of science and
industry to deal with uncertainties. Existing research in using fuzzy logic for the purpose of load
balancing has only concentrated in utilizing fuzzy logic concepts in describing processors load
and tasks execution length. The responsibility of the fuzzy-based load balancing process itself,
however, has not been discussed and in most reported work is assumed to be performed in a
distributed fashion by all nodes in the network. This paper proposes a new fuzzy dynamic load
balancing algorithm for homogenous distributed systems. The proposed algorithm utilizes fuzzy
logic in dealing with inaccurate load information, making load distribution decisions, and
maintaining overall system stability. In terms of control, we propose a new approach that specifies
how, when, and by which node the load balancing is implemented. Our approach is called
Centralized-But-Distributed (CBD). An evaluation study of the proposed algorithm shows that our
algorithm is able to reduce the average response time and average queue length as compared to
known load balancing algorithms reported in the literature.
Keywords: Dynamic Load Balancing, Fuzzy Logic, Distributed System, Algorithms.
1. INTRODUCTION
Various studies have shown that distributing the work load evenly among processors of a
distributed system highly improves system performance and increases resource utilization. Load
balancing in distributed computer systems may be defined as the process of redistributing the
work load among processors in the system to improve system performance [1]. Dynamic load
balancing algorithms monitor changes on the system work load and redistribute the work load
accordingly, e.g., [1]-[13]. A dynamic load balancing algorithm is usually composed of three
strategies: transfer strategy, location strategy, and information strategy. Transfer strategy decides
on which tasks are eligible for transfer to other nodes for processing. Location strategy nominates
a remote node to execute a transferred task. Information strategy is the information center of a
load balancing algorithm. It is responsible for providing location and transfer strategies at each
node with the necessary information required to take their decisions. Information strategy is an
important part of a load balancing algorithm. The worth and complexity of any dynamic load
balancing algorithm depends heavily on the performance of its information strategy. The
implementation responsibility or control of a dynamic load balancing algorithm can take three
different forms: centralized, distributed, or semi-distributed. In a centralized load distribution
algorithm, a single node (called central node) in the network is nominated to be responsible for all
load distribution in the network. In a distributed load balancing algorithm, the responsibility is
distributed where each node in the network carries an equal share of the responsibility and
1
An earlier version of this paper, [32], was presented at the International Conference on
Computer and Information Technology, Zurich, Switzerland, January 15-17, 2012.

Ali M. Alakeel
executes the same algorithm. In a semi-distributed load balancing algorithm, the network is
segmented into clusters where each cluster contains a set of nodes. The control within each
cluster is centralized, i.e. a central node is nominated to take charge of the load balancing
process within its set. In the semi-distributed approach, load balancing of the whole distributed
system is achieved through the cooperation of central nodes of each cluster, i.e. the responsibility
is distributed among the central nodes of each cluster.
This paper proposes a new fuzzy dynamic load balancing algorithm for homogenous distributed
systems. The proposed algorithm utilizes fuzzy logic, e.g., [29]-[31], in dealing with inaccurate
load information, making load distribution decisions, and maintaining overall system stability. In
terms of control, we propose a new approach that specifies how, when, and by which node the
load balancing is implemented. Our approach is called Centralized-But-Distributed (CBD). Also
we introduce a new location policy which utilizes Fuzzy Logic techniques in redistributing the load
from heavily loaded nodes to lightly loaded nodes in the system. Our load balancing algorithm
includes an explicit mechanism for monitoring and tuning system stability. System stability is
maintained through the dynamic adjustment and tuning of various parameters incorporated in the
algorithm.
The rest of this paper is organized as follows. Section 2 provides an overview dynamic load
balancing algorithms. Section 3 presents our proposed fuzzy dynamic load balancing algorithm.
An experimentation study of the proposed algorithm is presented in Section 4 and in Section 5,
we discuss our conclusions and future research.
2. AN OVERVIEW OF DYNAMIC LOAD BALANCING ALGORITHMS
This section provides an overview of dynamic load balancing in distributed systems with more
emphasis on the components related to our work presented in this paper. Specifically, in
subsection 2.1, different approaches of controlling the load balancing process is presented and
subsection 2.2 presents different load balancing strategies that we used in our evaluation study to
compare the performance of the proposed algorithm.
2.1 The Controlling Mechanism
Along with various load balancing strategies which may be applied independently or tailored to
enhance the performance of an algorithm for solving a certain problem, different policies of where
to put the control of the load balancing algorithm have been proposed in the literature:
centralized, distributed, or semi-distributed.
A centralized load balancing strategy assigns a single processor the responsibility of initiating and
monitoring the load balance operation. In this strategy, a dedicated processor gathers the global
information about the state of the system and assigns tasks to individual processors. Despite its
high potential of achieving optimal performance, centralized strategies have some disadvantages:
high vulnerability to failures, storage requirements for maintaining the state information -
especially for large systems, and the dependability of the performance of the system on the
central processor which could result in a bottleneck [1].
In a distributed load balancing strategy, each processor executes the same algorithm and
exchanges information with other processors about the state of the system. Each processor may
send or receive work on the basis of a sender-initiated or a receiver-initiated policy. In a sender-
initiated policy, the sender decides which job gets sent to which receiver. In a receiver-initiated
policy, the receiver searches for more work to do. Intuitively, queues are formed at senders if a
receiver-initiative policy is used, while they are formed at receivers if a sender-initiative policy is
used. Additionally, scheduling decisions are made when a new job arrives at the sender in a
sender-initiative, while they are made at the departure of a job in a receiver-initiative policy. The
determination of which policy is adopted depends upon the load transfer request which can be
initiated by an over-loaded or under-loaded processor [1], [3], [6], [7]. Assuming that process
migration cost under the two strategies is comparable, it has been demonstrated in, [4], [6], and

Ali M. Alakeel
[8], using analytical models and simulations, that sender-initiated strategies generally perform
better at lower system loads while receiver-initiated strategies perform better at higher system
loads. Some of the advantages offered by the distributed policy are: Fault tolerance, minimum
storage requirements to keep status information, and the availability of system state information
at all nodes. The distributed policy still has some disadvantage, one of which is that optimal
scheduling decisions are difficult to make because of the rapidly changing environment
introduced by the arrivals and departures from individual processors. Another disadvantage is the
extra communication overhead introduced by all processors trying to gather information about
each other. To mitigate this overhead, some distributed strategies minimize the amount of
information exchanged, which has a negative reflection on the performance of an algorithm.
The semi-distributed policy comes in the middle between centralized and distributed policies. It is
introduced to take the best of each and to avoid the major drawbacks of each of the two policies.
The semi-distributed strategy is based on the partitioning of the processors into equal sized sets.
Each set adopts a centralized policy where a central processor takes charge of load balancing
within its set. The sets together adopt a distributed policy where each central processor of each
set exchanges information with other central processors of other sets to achieve a global load
balance.
It has been shown in [1], that the semi-distributed policy produces a better performance than the
centralized and distributed policies. Research demonstrates that each central processor yields
optimal load balance locally within its set. Moreover, this policy does not incur high
communication overhead while gathering system state information. Although this policy is a
mediator between the centralized and the distributed ones, it fits large distributed systems better
than small systems.
2.2 Different Load Balance Strategies
This section identifies some, not necessarily all or the best, dynamic load balancing strategies
that have been reported in literature. It has to be emphasized that the selection of these specific
strategies is to serve our presentation only and not to be interpreted as an exhaustive selection or
classification.
2.2.1 The Random Strategy
Under a random load balancing strategy, a local node selects a remote node randomly and
transfers the job there for execution [5]. Upon receiving this job, the remote node executes it if its
load, i.e., queue length, is below a predefined threshold. Otherwise, this remote node will select a
new destination node for this job. To avoid having this job ping ponged among nodes without
getting serviced, a limit on the number of hops it could take is imposed which enforces the last
node to receive that job to execute the job when this limit is reached regardless of its current load.
As shown in [5], the performance of this simple strategy, which does not employ any information
in its selection, was significant as compared to a system with no load balancing at all. The
performance of this strategy is usually used as a reference point to compare other load balancing
algorithms that collect global information.
2.2.2 The Threshold Strategy
In this type of load balancing algorithms, a threshold value T is used to decide whether a task is
executed locally or remotely. The threshold is the processor's queue length. The queue length is
the number of processes in service plus the number of processes waiting in the queue. Threshold
value assumes static value in an implementation of the algorithm [3] and [6]. In this strategy, a
processor will try to execute a newly arriving task locally unless this processor threshold has been
reached. In this case, this processor will select another processor randomly and probe it to
determine if transferring the task to the probed processor will place it above the threshold or not.
If it does not, then the task is transferred to the probed processor which has to execute the task
there without any attempts to transfer it to a third one. If it does place it above the threshold, then
another processor is selected in the same manner. This operation continues up to a certain limit
called the probing limit. After that, if no destination processor is found, the task is executed locally

Ali M. Alakeel
[6]. It should be noted that the threshold strategy requires no exchange of information among
processors in order to decide whether to transfer a task or not [6]. This is an advantage because
it minimizes communication overhead. Another advantage is that the threshold policy avoids extra
transfer of tasks to processors that are above the threshold. It has been shown, in [6], that the
threshold algorithm with T=1 yields the best performance for low to moderate system load, and
the threshold algorithm with T=2 gives the best performance for high system load.
2.2.3 The Greedy Strategy
This strategy, the current state of each processor P represented by a function f(n) , where n is the
number of tasks currently at the processor. If a task arrives at P and number of tasks n is greater
than zero, then this processor looks for a remote processor that has its state less than or equal to
f(n). If a remote processor is found with this property, then the task is transferred there. The
performance of this strategy depends on the selection of the function f(n). It has been shown in
[33] that f(n) < n must holds in order to achieve good performance. Also, n-1, n div 2, n div 3, n
div 4, etc. are possible values for f(n). Furthermore, it has been shown in [33] that f(n) = n div 3
yields the best results and that the greedy strategy outperforms the threshold strategy with T=1 in
all experiments. The greedy strategy adopts a cyclic probing mechanism instead of the random
selection used in the threshold strategy. In this cyclic probing mechanism, processor i probes
processor (i+j) mod N, N representing the number of processors in the system, in the j
th
probe to
locate a suitable destination processor. For example, in a system with 5 processors numbered
0,1,2,3, and 4 respectively, Processor 1 will first probe processor 2. If this attempt is not
successful, it will probe 3 and so on. As in the threshold strategy, once a task is transferred to a
remote processor it must be executed there [33]. Despite the similarities between the two
strategies, it has been demonstrated using simulation results in [33] that the greedy strategy
outperforms the threshold strategy. This improvement is attributed to the fact that the greedy
strategy attempts to transfer every task that arrives at a busy processor whereas the threshold
strategy attempts to transfer only when a task arrives at a processor which has reached the
threshold T or higher.
2.2.4 The Shortest Queue Strategy
The shortest load balancing strategy selects a subset of remote nodes randomly and probes
them to find out their current load, i.e. queue length. The remote node with the smallest queue
length is then selected. If this selected node’s queue length is smaller than a certain threshold,
then the job is transferred there, otherwise the job is executed locally. Although the shortest
strategy attempts to make a wiser selection than the threshold does, it is shown in [5] that there
was not a significant gain in performance over what has been achieved by the threshold strategy.
2.2.5 The Bidding Strategy
The basic idea of this strategy is bids. The overloaded processor looking for help in executing
some of its tasks requests other processors to submit their bids. Bid information includes the
current work load status about each processor. After receiving all bids from participating
processors, the original processor selects to whom it will send some of its tasks for execution. A
major drawback of this strategy is the possibility that one processor will become overloaded as a
result of its winning many bids. To overcome this problem, some variations of this strategy would
allow the bidder processor to accept or reject the tasks sent by the original processor. This could
be done by allowing it to send a message to the original processor informing it of whether the
work has been accepted or rejected. Since a processor's load could change while these
messages take place, the final selection might not turn out to be as good as it seems to be at
earlier time or vice versa [7]. Different algorithms have been proposed in the literature to
determine who gets to initiate the bid, bid information, bid selection, bid participation, and bid
evaluation [1], [3], [4], [5], [7]. The performance of this strategy depends on the amount of
information exchanged, the selection of bids, and communication overhead [3]. More information
exchange enhances the performance and provides a stronger basis for selection but also
requires extra communication overhead [3], [7], [33].

Ali M. Alakeel
2.2.6 The Drafting Strategy
The drafting strategy differs from the bidding strategy in the way it allows process migration and
in the manner it attempts to achieve load balance. The drafting policy tries to alleviate some of
the communication overhead introduced by the bidding strategy. The drafting policy achieves
load balance by keeping all processors busy rather than evenly distributing the work load among
participating processors (which is one of the objectives of the bidding strategy). In the bidding
strategy, to keep all processors evenly loaded, groups of processes will be required to migrate
from a heavily loaded processor to a lightly loaded processor. Consequently, it is possible to find
some of these processes migrating back as a result of the unpredictable change of the
processor's work load. To allow for this problem in the bidding approach, the drafting strategy
allows only one process to migrate at a time rather than group migration [7].
The drafting strategy adopts a process migration policy which is based on giving the control to the
lightly loaded processors. Lightly loaded processors initiate a process migration instead of having
process migration being triggered by a processor being overloaded as in the bidding strategy [7].
In drafting, the number of processes currently at the processor is used for work load evaluation.
Each processor maintains its work load and identifies itself as in one of the following states: H-
Load, N-Load, or L-load. An H-Load, heavy load, indicates that some of this processor's
processes can migrate to other processors. An N-Load, normal load, indicates that there is no
intention for process migration. A L-Load, light-load, indicates that this processor is willing to
accept some migrant processes. A load table is used at each processor to hold this information
about others processors and act as a billboard from which the global information of the system is
obtained. When a load change occurs in a processor, it will broadcast its load information to other
nodes so that they will update their load tables.
When the processor becomes lightly loaded, i.e. L-Load, it will identify other processors having
the status of H-Load from its load table and send them a draft-request message. This message
indicates that the drafting processor is willing to accept more work. If by the time it receives this
message it is still in H-Load, each remote (drafted) processor will respond by sending a draft-
respond message which contains draft-age information. Otherwise the current load status will be
returned to the drafting processor, adopting the concept that a process is allowed to migrate only
if it is expecting a better response time and age is associated with each draftable process. Some
of the parameters that may be used for age determination are: Process waiting time, process
priority, or process arrival [7]. The draft-age is determined by the ages of those processes
nominated to be drafted. Various alternatives for draft-age calculations are possible. The
selection of the draft age to be the maximum age of all draftable processes, the average age of
the draftable processes, or simply the number of draftable processes are some of them [7]. When
all draft-response messages are received, the drafting processor calculates draft-standard
criteria. Draft-standard criteria are calculated based on the draft-ages received and used to
ensure fairness of selection among drafted processes. The choice of draft standard is crucial to
the performance of this strategy and is determined at the system design stage. After calculating
the draft-standard, a draft-select message is sent to the drafted processor that has the highest
draft-age. The drafting processor will send the draft-select message, only if it is still on the L-Load
state, otherwise it will not accept any migrating processes.
3. THE PROPOSED FUZZY LOAD BALANCING ALGORITHM
In this section, we present our proposed load balancing algorithm for homogeneous distributed
computer systems that is based on fuzzy logic capabilities.
3.1 A Background
In our daily life, we use words and terms, which are vague or fuzzy such as:
“The server is slow” or
“The weather is hot” or
“John is tall.”

Ali M. Alakeel
Fuzzy Logic [29], [31], gives us the ability to quantify and reason with words which have
ambiguous meanings such the words above. In fuzzy sets [29], an object may belong partially to
a set as opposed to classical or “crisp” sets in which an object may belong to a set or not. For
example, in a universe of heights (in feet) for adult people defined as:
µ= {5, 5.5, 6, 6.5, 7, 7.5, 8}, a fuzzy subset TALL can be defined as follows:
TALL = [0/5, .125/5.5, .5/6, .875/6.5, 1/7, 1/7.5, 1/8].
In this simple example, the degree of membership for the members of the universe µ, with
respect to the set TALL may be interpreted as that the value “6” belongs to the set TALL 60%
percent of the time while the value 8 belongs to the set TALL all the time.
3.2 The System Model
In this presentation, we assume the following system model. We assume N, where N > 1,
independent homogenous nodes are connected by a local area network where each node
consisting of a single processor. A single typed tasks arrive to a node could either come from
outside the network or from other nodes in the network. We assume that all nodes are subjected
to the same average arrival rate of tasks coming from outside the network. All tasks are queued
at each node and are served on a First Come First Serve (FCFS) basis.
3.3 The Assumptions
Our presentation is based on the following assumptions:
 Nodes of the distributed system are numbered from 1 to N, where N is the total number
of nodes in the system. The numerical number of each node represents its identification
(ID) in the system.
 Nodes are connected by a broadcast network and the cost of sending a message
between any two nodes is the same.
 We assume that each processor is in some state Sk where it has a number of tasks Tk.
Furthermore, we assume that the distributed system is initially in a steady state.
 Given this configuration, the load balancer starts and tries to examine the overall state of
the system and takes the necessary corrective actions accordingly based on the
objectives this algorithm assumes.
3.4 The Objectives
In this presentation, once a node is elected, by the algorithm, to act as the load balancer of the
distributed system under consideration, this node will have the following objectives:
1) To obtain full information about the load of each host in the system efficiently.
2) To maintain a load balance among the distributed system hosts within a difference d.
The value of d varies during the operation of the load balancing algorithm and adjusts
dynamically taking into account the current state of the system and the communication
costs. The proper range of d will only be determined after experimentations with this
algorithm because this process has to be done efficiently in terms of communication time
required.
3) To select the most appropriate time to launch the load balancing process. Since keeping
the load balancer working at all times is a burden on the system, we propose an efficient
way of triggering the load balancer, which is described as follows. The load balancer
algorithm is triggered when one of the following conditions is satisfied:
a) When a node becomes idle, or below a threshold value.
b) When the load of a node exceeds a threshold value.
4) To ensure that only one node is working as the load balancer at a time. It is possible that
more than one node can meet either one of the above conditions. This would cause
multiple load balancers to be active at the same time. To prevent this, our algorithm
provides a mechanism which ensures that only one load balancer is active at a given
time.

Ali M. Alakeel
3.5 The Algorithm Steps
According to the proposed fuzzy based load balancing algorithm, the load balancer node
performs the following steps:
1) Obtain the current load of the distributed system. This is accomplished by broadcasting a
request for status message to all nodes in the system.
2) Upon receiving the response from all nodes, the load balancer assigns each node of the
system including itself a fuzzy value in the interval [0,1] which represents the load of that
node relative to the overall load of the distributed system. This assignment is achieved by
forming a fuzzy set, LOADED = {light, normal, heavy}, that represents the load of the
system. Using Fuzzy Logic techniques each node of the distributed system is assigned a
membership value depending on its current load. The membership value is in the interval
[0,1] and reflects the compatibility of the load at a specific node to the fuzzy term
LOADED which is represented by a fuzzy set. The assignment of membership values
(grades) is based on the S-function [31] which is shown in Fig. 1. S-functions may be
described as follows.
 A mathematical function that is used in fuzzy sets as a membership function.
 A simple but valuable tool in defining fuzzy functions such as the word “load.”
 The symbols, , and  are parameters which may be adjusted to fit the
desired membership data.
 The S-function is flat at a value of zero for × ≤  and at 1 for × ≥ . In between
 and , the S-function is a quadratic function of ×. The  parameter
corresponds to the crossover point of 0.5 and is (  ) / 2.
3) Using the results of step (2), the load balancer categorizes all nodes into three separate
groups: Underloaded, Normal, and Overloaded. Where Normal nodes do not need any
load balancing and will be left unchanged. Both Underloaded and Overloaded nodes
need help and will be the target of the load balancing process.
4) Create a mapping from Overloaded nodes to Underloaded nodes. The outcome of this
mapping tells each overloaded node where to ship some of the extra work it has. As a
result of this mapping, the load balancer sends each overloaded node a message
specifying the ID of each possible underloaded node and the number of tasks the
overloaded node should ship to the underloaded node. This information is formed in a
list arranged as: (ID1, #tasks), (ID2, #tasks),…, (IDN, #tasks). To perform this mapping, we
adopt Evans et al. [31] probability model by using the load at each node and compute the
probability of sending a task from an overloaded node i to and underloaded node j.
3.6 An Illustrative Example
In this section, we present a simplified example that shows how system load may be dealt with
using fuzzy logic in our proposed algorithm.
▫ Assume a distributed system with 4 nodes, labeled as N1, N2, N3 and N4.
▫ Assume that at a time t, the average queue length in the system is 15 jobs.
▫ Let  = {1, 2, 3, 4, …, 15} be the universe of jobs.
▫ To describe our system load using fuzzy logic, we can create three different fuzzy sets
[29] as follows:
▫ lightly-loaded = {1/1, .75/2, .5/3, .25/4} from the codomain l ={1, 2, 3, 4}
▫ normally-loaded= {1/4, .75/5, .5/6, .25/7} from the codomain n ={4, 5, 6, 7}, and
▫ heavily-loaded = { .2/7, .3/8, .4/9, .5/10, .6/11, .7/12, .8/13, .9/14, 1/15} from the
codomain h ={7, 8, 9, 10, 11, 12, 13, 14, 15}.
▫ Assume that at a time t, the load at each node is found to be as follows:
▫ N1 (8 jobs), N2 (5 jobs), N3 (11 jobs) and N4 (2 jobs).
▫ A possible mapping of nodes to the three fuzzy sets is:
▫ L_load = {0/1, .5/2, 0/3, .9/4},
▫ N_load = {.125/1, .75/2, 0/3, .25/4}, and

Ali M. Alakeel
▫ H_load = {0.5/1, 0/2, .875/3, 0/4}.
It should be noted that the main contribution of our proposed algorithm is in the part which is
responsible for selecting the best time to trigger the load balancing process and in performing the
load balancing process itself. In this regard, a copy of the load balancing algorithm resides on
each host as in the distributed load balancing algorithm reported in the literature. In our
approach, the responsibility of load balancing is centralized in action, but distributed in time. This
means that only a single node will be responsible for performing the load balancing processes of
the distributed system. This node, however, is not the only one who will complete this task all of
the time, as is in the traditional centralized approach. Instead, each node might have the chance
to perform the load balancing task. The selection of which node actually does the work is
dynamically determined, however. By doing this, our approach attempts to get the best of the
well-known centralized and distributed approaches to the load balancing problem.
FIGURE 1: The S-function.
4. AN EXPERIMENTAL STUDY
The In order to evaluate the effectiveness of our proposed fuzzy load balancing algorithm for
homogeneous distributed systems, we have conducted an evaluation study using simulation. In
this experiment, we have assumed that job’s inter-arrival time follows the exponential distribution.
Also, we have assumed that job’s execution time is drawn from the uniform distribution.
Furthermore, we have assumed a fixed communication time between each two hops in the
network. In our experiment, we have used a network composed of four nodes of the same
configuration. In this study, a node may get load (jobs) from outside the network or from inside
the network because a job inside the network is allowed to move from one node to another for a
service. During the period of the experiment, we have subjected each node to different surges of
requests from outside of the network.

Ali M. Alakeel
FIGURE 2: The Performance comparison based on average response time.
FIGURE 3: The Performance comparison based on average queue length.
At a certain periods of time, the load balancing algorithm is triggered as described in Sec 3.
Based on the decisions made by the load balancing algorithm, the load (jobs) may move from
node to another in order to load-balance the network. In this stage of our evaluation study, we
have measured two major performance indicators of distributed systems, namely, the average
response time and the average processor’s queue length. We have compared our algorithm to a
well-known load balancing strategies from those discussed in Sec. 2. In Fig. 2, the x-axis projects

Ali M. Alakeel
the system load represented by a fuzzy set as described in Sec. 3, while the y-axis reflects the
system’s average response time in milliseconds.
As shown in Fig. 2, the proposed fuzzy load balancing algorithm, “Fuzzy,” was able to reduce the
average response time as compared to Threshold, Shortest and Random load balancing
strategies. In Fig. 3, the x-axis projects the average queue length at each processor in our
system, while the y-axis reflects the system’s average response time in milliseconds. As shown in
Fig. 3, the proposed algorithm was also able to reduce job’s queue length as compared to the
Random and Threshold strategies. Although the results of this experiment are encouraging, the
proposed algorithm still needs to be evaluated against different systems load and larger
distributed system with thousands of processors.
5. CONCLUSIONS AND FUTURE RESEARCH
This paper proposes a new dynamic load balancing algorithm for homogenous distributed
computer systems which employs fuzzy logic in dealing with inaccurate load information, making
load distribution decisions, and maintaining overall system stability. The main contribution of our
proposal is in the part which is responsible for selecting the best time to trigger the load balancing
process and in performing the load balancing process itself. In order to evaluate the proposed
algorithm, an overvaluation study was conducted using simulation. The result of this experimental
study shows that the proposed fuzzy load balancing algorithm was able to produce a promising
system’s performance improvements as compared to well-known balancing algorithms reported
in the literature. For future research, it is imperative to conduct more experiments that evaluate
the performance of the proposed model in computer networks consisting of thousands of nodes.
Additionally, it is our intention to construct a theoretical model for our proposed algorithm and
compare the results reached with our simulation results presented in this paper.
6. REFERENCES
[1] I. Ahmed and A. Ghafoor, "Semi-Distributed Load Balancing for Massively Parallel
Multicomputers," IEEE Trans. Software Eng., vol. 17, no. 10, pp 987-1004, October 1991.
[2] T. L. Casavant, "A Taxonomy of Scheduling in General-Purpose Distributed Computing
Systems," IEEE Trans. Software Eng., vol 14, no. 2, pp 141-154, February 1988.
[3] Y. Wang and R. Morris, "Load Sharing in Distributed Systems," IEEE Trans. Comput., vol.
C-34, no. 3, pp. 204-217, Mar. 1985.
[4] K. Ramamritham, J. A. Stankovic, and W. Zhao, "Distributed Scheduling of Tasks with
Deadlines and Resource Requirements," IEEE Trans. Comput., vol. 38, no. 8, pp 1110-
1123, August 1989.
[5] J. A. Stankovic, K. Ramamritham, and S. Cheng, "Evaluation of a Flexible Task Scheduling
Algorithm for Distributed Hard Real-Time Systems," IEEE Trans. Comput., vol. C-34, no. 12,
pp. 1130-1143, December 1985.
[6] D.L. Eager, E.D. Lazowski, and J. Zahorjan, "Adaptive Load Sharing in Homogeneous
Distributed Systems," IEEE Trans. Software Eng., vol. SE-12, no. 5, pp. 662-675, May 1986.
[7] L. M. Ni, C. Xu, and T. B. Gendreau, "A Distributed Drafting Algorithm for Load Balancing,"
IEEE Trans. Software Eng., vol.SE-11, no. 10, pp. 1153-1161, October 1985.
[8] D. L. Eager and E. D. Lazowski, and J. Zahorjan, "A Comparision of Receiver-Inititated and
Sender Initiated Adaptive Load Sharing," Performance Evaluation, 6, pp. 53-68, March,
1986.

Ali M. Alakeel
[9] G. Cybenko, “Dynamic load balancing for distributed memory multiprocessors,” J. Parallel
Distrib. Comput. 7 , pp. 279–301, 1989.
[10] J. Watts, S. Taylor, “A practical approach to dynamic load balancing,” IEEE Trans. Parallel
Distrib. Systems 9 (3), pp. 235–248, March 1998.
[11] P. Krueger, N.G. Shivaratri, “Adaptive location policies for global scheduling,” IEEE Trans.
Software Eng. 20 (6), pp. 432-444, June 1994.
[12] S. Dhakal, M. M. Hayat, J.E.Pezoa, C. Yang, and D. Bader, "Dyanmic Load Balancing in
Distributed System in the Presence of Delays: A Regeneration-Therory Approach," IEEE
Transactions on Parallel and Distributed Systems, vol. 18, no. 4, April 2007.
[13] D. J. Evans and W.U.N. Butt,” Dynamic load balancing using task-transfer probabilities,”
Parallel Computing, Vol. 19, No. 8, pp. 897-916, August 1993.
[14] A. M. Alakeel, “Load Balancing in Distributed Computer Systems,” Int. Journal of Computer
Science and Information Security, Vol. 8, No. 4, pp. 8-13, July 2010.
[15] A. M. Alakeel, “A Guide to Load Balancing in Distributed Computer Systems,” Int. Journal of
Computer Science and Network Security, Vol. 10, No. 6, pp. 153-160, June 2010.
[16] K. Abini, “Fuzzy Decision Making for Load Balancing in a Distributed System, “ Proceedings
of the 36th Midwest Symposium Circuits and Systems, pp. 500–502, 1993.
[17] Yu-Kwong Kwok, Lap-Sun Cheung, “A new fuzzy-decision based load balancing system for
distributed object computing,” Journal of Parallel and Distributed Computing, Volume 64,
Issue 2, pp. 238-253, February 2004.
[18] L. Singh, A. Narayan, and S. Kumar, "Dynamic fuzzy load balancing on LAM/MPI clusters
with applications in parallel master-slave implementations of an evolutionary neuro-fuzzy
learning system," IEEE International Conference on Fuzzy Systems, pp.1782-1788, June
2008.
[19] C.W. Cheong, V. Ramachandran, “Genetic Based Web Cluster Dynamic Load Balancing in
Fuzzy Environment,” Proceedings of the Fourth International Conference on High
Performance Computing in the Asia-Pacific Region, Beijing, China, Vol. 2, pp. 714–719,
2000.
[20] P. Chulhye, J.G. Kuhl, “A fuzzy-based distributed load balancing algorithm for large
distributed systems,” Proceedings of the Second International Symposium on Autonomous
Decentralized Systems, pp. 266–273, April 1995.
[21] M. Rantonen, T. Frantti, and K. Leiviskä, "Fuzzy expert system for load balancing in
symmetric multiprocessor systems," Journal of Expert Systems with Applications, Vol. 37
No. 12, pp. 8711-8720, December, 2010.
[22] L. Singh, A. Narayan, and S. Kumar, "Dynamic fuzzy load balancing on LAM/MPI clusters
with applications in parallel master-slave implementations of an evolutionary neuro-fuzzy
learning system," IEEE International Conference on Fuzzy Systems, pp.1782-1788, June
2008.
[23] I. Barazandeh, S. S. Mortazavi, and A. M. Rahmani, "Intelligent fuzzy based biasing load
balancing algorithm in distributed systems," IEEE 9th Malaysia International Conference,
pp.713-718, Dec. 2009.

Ali M. Alakeel
[24] E. El-Abd, “Load blancing in distrubuted computing systesms using fuzzy sexpert ssytems, “
Int. Confernce on Modern Probmesn of Radio Engiennering, Telecommunications and
Computer Scinece, Lviv-Slavsko, Ukraine, pp. 141-144, 2000.
[25] S. Dierkes, “Load balancing with a fuzzy-decision algorithm,” Inform. Sci. 97 (1–2), pp. 159–
177, March 1997.
[26] S. P. McAuliffe, “Job scheduling using fuzzy load balancing in distributed system,” Electron.
Lett. 34 (20), pp. 1983–1985, October 1998.
[27] K.-W.Wong, “Fuzzy routing control of service request messages in an individual computing
environment,” Proceedings of ACM Symposium on Applied Computing, Nashville, TN, pp.
548–551, 1995.
[28] A. Kumar, M. Singhal, and Ming T(M&e) Liu, "A Model for Distributed Decision Making: An
Expert System for Load Balancing in Distributed Systems", IEEE computer software and
applications conference, 1987.
[29] L. A. Zadeh, “Fuzzy Sets,” Information and Control, No. 8, pp. 338-353, 1965.
[30] B. Kosko, “Neural Networks and Fuzzy Systems: A Dynamical Systems Approach to
Machine Intelligence,” Prentice-Hall, Englewood Cliffs, NJ, 1992.
[31] J. Giarratano, “Expert Systems: Principles and Programming,” PWS-KENT Publishing
Company, Boston, 1989.
[32] A. M. Alakeel, "A Fuzzy Dynamic Load Balancing Algorithm for Homogenous Distributed
Systems,” Proceedings of the Int. Conference on Computer and Information Technology,
Zurich, Switzerland, pp. 63-66, January 2012.
[33] S. Chowdhury, "The Greedy Load Sharing Algorithms," J. Parallel and Distributed Comput,
vol. 9, pp. 93-99, May 1990.

Application of Fuzzy Logic in Load Balancing of Homogenous Distributed Systems1

Recommended

More Related Content

What's hot (20)

Viewers also liked (20)

Similar to Application of Fuzzy Logic in Load Balancing of Homogenous Distributed Systems1 (20)

Recently uploaded (20)

Application of Fuzzy Logic in Load Balancing of Homogenous Distributed Systems1