A SURVEY ON STATIC AND DYNAMIC LOAD BALANCING ALGORITHMS FOR DISTRIBUTED MULTICORE SYSTEM
0 likes8 views
This document summarizes a survey of static and dynamic load balancing algorithms for distributed multicore systems. It discusses how efficient load balancing is essential for distributing work across cores in large supercomputers. Both static and dynamic algorithms are reviewed. Static algorithms allocate work deterministically or probabilistically without considering runtime conditions, while dynamic algorithms can adapt based on network conditions and core capabilities. The paper evaluates various performance metrics for different load balancing algorithms and concludes that modern distributed multicore systems require more reliable dynamic algorithms to optimize performance.
![International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 09 Issue: 06 | June 2022 www.irjet.net p-ISSN: 2395-0072
© 2022, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page 2535
A SURVEY ON STATIC AND DYNAMIC
LOAD BALANCING ALGORITHMS FOR DISTRIBUTED MULTICORE
SYSTEM
Dr.G.Muneeswari1
1Professor, School of Computer Science and Engineering, VIT-AP University, Amaravati, Andhra Pradesh, India
---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - In today’s world with large number of super
computers, it’s essential to distribute the load efficiently
among different cores among the multicore processing with
proper load balancing algorithms. The core idea is that none
of the multicore elements should be kept in the idle state. In-
order to implement the efficient algorithm, the better
understanding of the distributed system must be dynamically
known which actually a challenging task is. Modern
framework for load balancing methodologies are integrated
with the application itself whichcanbalancetheloadbasedon
the application level knowledge. But the real scenario lies in
the work of processing elements capabilities. So, these
algorithms may lead to incorrect simulations. This paper
focusses on the comprehensivestudyofexistingload balancing
algorithms implemented on the various distributed
environments.
Key Words: Load balancing, static, dynamic, resource
utilization, Multicore, distributed system
1. INTRODUCTION
The distributed multicore system can be homogeneous or
heterogeneous with different architectures. A process
allocated to the multicore processing elements can be pre-
emptive or non-pre-emptive. A proper and efficient load
balancing algorithm must be incorporatedmostlyinthecase
of homogeneous systemfor betterresourceutilization.Inthe
traditional system mostly sender initiated and the receiver
initiated algorithms are implemented which may not be
useful for the current massive parallel technology. A new
novel hybrid algorithms which can combine the feature of
static and dynamic allocation is required for keeping all the
processing elements in the busy state.
The load can be divided based on the implementation of the
algorithm. Ideally, underload, overload thresholds are fixed
at the centralized master node depends on the expected
number of process arrival and capability of computational
elements speed of execution. In this paper, a two variant of
static and dynamic algorithms are reviewed for various
cloud based multicore environments.
The organization of the paper is given as follows: Section 2
describes the literature review and section 3 explains the
performance metrics associated with the load balancing
algorithms. Finally section 4 concludes the paper.
2. LITERATURE REVIEW
In this method, the performance (Motwani & Raghavan [1],
Zhong Xu & Rong Huang [2] of all the processing elements
are calculated before starting any task. This distributed
multicore system follows the strategy of master slave
processing. The master node allocates the task to various
nodes based on the estimated load and arrival time of the
tasks to the ready queue. Sometimes the slave nodes will be
uded to calculate the work load of the incoming requests. In
the McEntire et al [3] algorithm, a designated computational
element selects a site for the execution of the new task.
Whenever the task is created, a lightly loaded processing
element is chosen depending on the complete picture of
workload. Depending on the tasteful information, the load
balancing algorithm quicklydecides upon the loadallocation
and migration.
Unlike the static kind of algorithms, the dynamic algorithms
proposed by Malik[4], Wang & Morris[5] allocates the tasks
dynamically when one of the processors becomes under-
loaded. In [6], the routing for hybrid wireless network
becomes complicated with respect high transmission rate.
When the intermediate hops are preoccupied with data then
the load balancing algorithm find out a different route based
on the estimated data load during data transfer. The load
balancing [7] forcloud willbeextremelycomplexasthereare
multiple nodes requested at the same time for client data
access. A novel methodology related to the workload
distribution is implemented on the RAID systems for the
better throughput and performance.
IOT systems[8] have less memory and less power active
devices wherein the real challenge is to find out the load of
data transfer in web based IOT application. Data transfer to
the web servers are implemented with the help oftraditional
sender oriented load balancing algorithms. Mohammed et
al.,[9] proposed a server consolidation methodology for](https://image.slidesharecdn.com/irjet-v9i6466-221017063717-ab47a670/85/A-SURVEY-ON-STATIC-AND-DYNAMIC-LOAD-BALANCING-ALGORITHMS-FOR-DISTRIBUTED-MULTICORE-SYSTEM-1-320.jpg)
![International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 09 Issue: 06 | June 2022 www.irjet.net p-ISSN: 2395-0072
© 2022, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page 2536
incorporating load balancing algorithms in cloud based
environments. Cloud based datacenters are taken into
consideration with master slave activation and the master
nodeor host is responsible for data transferforbalancingthe
load in the distributed environment. The overall survey [10]
of load balancing algorithm along with the different
performance metric comparison has been depicted in the
work of Kashyap et al.,Mainly the distributed file system
architecture is evaluated for the advanced cloud based
systems. Central queue algorithm proposed by William
Leinberger [11] based on the working of simple FIFO
mechanism.Wheneverthereisresourcerequestrequirement
comes to the centralized node which has the queue, it will be
inserted at the beginning of the queue and whenever the
processor becomes available then the request will be
removed and given for execution.
Similarly, in-order to implement the load balancing concept,
the task migration can be done in FIFO queue order. In
homogeneous system [11], the real challenge predominantly
lies in the allocation of requests to the respective servers
where the data is present. Resources are placed in aresource
pool and the resource manager is responsible for handling
the incoming requests. Parallel computing system [12]
handlesload balancing with the help of master nodesandthe
slave nodes will be involved only with the computational
work.Tremendousamountofmassiveparallelismisachieved
with the help of this novel idea.
3. EVALUATION PAPRAMETERS
The performance of various load balancing algorithms is
measured by the following parameters.
Eliminating overload
Whenever there is a overload situation which cannot be
handled by the existing algorithms then the load balancing
algorithmsstopsexecutionandeliminatesfurtheracceptance
of workload from the clients.
Resource availability
This parameter will be very crucial for the user level
satisfaction as the user always wantsa seamless executionof
tasks for their different requests. Availability always related
to the fault tolerance of the distributed multicore systems.
Fault tolerance makes the system work even in the case of
failure of the system. So for better performance of workload
migration fault tolerance is very important.
Accuracy level Identification
Static load balancing algorithms provide more accuracywith
respect to the identification of workloadasitispredefined.In
the dynamic algorithms the speculation of work load
sometimes depends on the delays inculcated on the network
and leads to incorrect prediction of load.
Single Point of Access Vs Distributed
In the single point load balancing algorithms every load
balancing decision as well as task migration will be
happening at only one centralized node and the rest of the
nodes or sites will be actively participating in the
computational work. Ideally in thedistributedloadbalancing
algorithm all the nodes will be involved in the load balancing
decision making which makes the entiresystemafairkindog
balancing the work load.
Dependent and Independent Processing
The decision of load migration depends on the type of
execution carried on the processing elements on the shared
system. If a task is divided into several subtasks and each
individual subtasks are carried on different geographically
distributedsystemsthenthemigrationofworkloadshouldbe
carefully done by the load balancing algorithm for the final
completion of the original task. For independent tasks, there
will not be any complication as they can independently
complete the execution.
Hardware and Software Utilization
Resource utilization request from the clientcan be hardware
or software oriented. If more number of utilization comes
from the same client and if other client requests are not
handled then more adaptable dynamic strategies can be
implemented.
4. CONCLUSION
Basically load balancing algorithms are categorizedeitheras
a static algorithm or dynamic algorithm. In this paper, we
reviewed mostly static load balancing algorithms where it
allocates the work to the processing elements
probabilistically or deterministically and do not considerthe
application level runtime scenarios. It is concluded that it is
not possible to find out the time of arrival of the process and
also speculating the future execution times of different
workloads. We also investigated very few dynamic
algorithms in which the load can be predictedwithrespectto
the network condition and processing elements capabilities.
In modern technology implemented on top of the distributed
multicore systems, a more reliable dynamic algorithms can
be incorporated for better performance.
REFERENCES
[1] Motwani, R & Raghavan, P 1996, ‘Randomized
algorithms’, ACM Computing Surveys (CSUR), vol.28(1),
pp.33-37.](https://image.slidesharecdn.com/irjet-v9i6466-221017063717-ab47a670/85/A-SURVEY-ON-STATIC-AND-DYNAMIC-LOAD-BALANCING-ALGORITHMS-FOR-DISTRIBUTED-MULTICORE-SYSTEM-2-320.jpg)
![International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 09 Issue: 06 | June 2022 www.irjet.net p-ISSN: 2395-0072
© 2022, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page 2537
[2] Zhong Xu & Rong Huang 2006, ‘Performance Study of
Load Balancing Algorithms in Distributed Web Server
System’, CS213 Parallel and Distributed Processing
Project Report.
McEntire Munro D,S & Brown, A,L 2000, ‘Evaluating
partition selection policies using the CMOS garbage
collector’, POSG ninth International workshop on
Persistent Objects Systems, pp. 104-115.
[3] Malik, S 2000, ‘Dynamic Load Balancing in a Network of
Workstation’, 95.515 Research Report.
[4] Wang, Z & O’Boyle, MF 2009, ‘Mapping parallelism to
multi-cores: a machinelearningbasedapproach’Proc.of
the 14th ACM SIGPLAN Symposium on Principles and
Practice of Parallel Programming.
[5] Yan Yan, Qiang Hu, Douglas M. Blough, "Load-Balanced
Routing for Hybrid Fiber/WirelessBackhaul Networks",
2021 IEEE Global Communications Conference
(GLOBECOM), pp.1-6, 2021.
[6] N. S. Raghava and Singh Deepti, "Comparative Study on
Load Balancing Techniques in Cloud Computing", Open
journal of mobile computing and cloud computing, vol. 1,
no. 1, pp. 18-25, August 2014.
[7] Sagar Vats, Yachana Kumari, Tushar Singh, Sunil
Chandra, Sushma Shirke, "LoadBalancinginSpecificIOT
Systems", 2021 International Conference on Intelligent
Technologies (CONIT), pp.1-6, 2021.
[8] Mohammed Ala’anzy, Mohamed Othman, "Load
Balancing and Server Consolidation inCloudComputing
Environments: A Meta-Study", IEEE Access, vol.7,
pp.141868-141887, 2019.
[9] Kashyap Dharmesh and Viradiya Jaydeep, "A Survey Of
Various Load Balancing Algorithms in Cloud
Computing", International Journal Of Scientific &
Technology Research, vol. 3, no. 11, pp. 115-119,
November 2014.
[10] WilliamLeinberger,George Karypis&Vipin Kumar2000,
‘Load Balancing Across Near-Homogeneous Multi-
Resource Servers’, 0-7695-0556-2/00, IEEE.
[11] Archer C.J., Mullins T.J, Sidelnik A. and Smith B.E.
“Parallel Computing System Using Coordinator and
Master Nodes for Load Balancing and Distributing
Work”. United State Patent, 2010.](https://image.slidesharecdn.com/irjet-v9i6466-221017063717-ab47a670/85/A-SURVEY-ON-STATIC-AND-DYNAMIC-LOAD-BALANCING-ALGORITHMS-FOR-DISTRIBUTED-MULTICORE-SYSTEM-3-320.jpg)
A SURVEY ON STATIC AND DYNAMIC LOAD BALANCING ALGORITHMS FOR DISTRIBUTED MULTICORE SYSTEM
- 1. International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056 Volume: 09 Issue: 06 | June 2022 www.irjet.net p-ISSN: 2395-0072 © 2022, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page 2535 A SURVEY ON STATIC AND DYNAMIC LOAD BALANCING ALGORITHMS FOR DISTRIBUTED MULTICORE SYSTEM Dr.G.Muneeswari1 1Professor, School of Computer Science and Engineering, VIT-AP University, Amaravati, Andhra Pradesh, India ---------------------------------------------------------------------***--------------------------------------------------------------------- Abstract - In today’s world with large number of super computers, it’s essential to distribute the load efficiently among different cores among the multicore processing with proper load balancing algorithms. The core idea is that none of the multicore elements should be kept in the idle state. In- order to implement the efficient algorithm, the better understanding of the distributed system must be dynamically known which actually a challenging task is. Modern framework for load balancing methodologies are integrated with the application itself whichcanbalancetheloadbasedon the application level knowledge. But the real scenario lies in the work of processing elements capabilities. So, these algorithms may lead to incorrect simulations. This paper focusses on the comprehensivestudyofexistingload balancing algorithms implemented on the various distributed environments. Key Words: Load balancing, static, dynamic, resource utilization, Multicore, distributed system 1. INTRODUCTION The distributed multicore system can be homogeneous or heterogeneous with different architectures. A process allocated to the multicore processing elements can be pre- emptive or non-pre-emptive. A proper and efficient load balancing algorithm must be incorporatedmostlyinthecase of homogeneous systemfor betterresourceutilization.Inthe traditional system mostly sender initiated and the receiver initiated algorithms are implemented which may not be useful for the current massive parallel technology. A new novel hybrid algorithms which can combine the feature of static and dynamic allocation is required for keeping all the processing elements in the busy state. The load can be divided based on the implementation of the algorithm. Ideally, underload, overload thresholds are fixed at the centralized master node depends on the expected number of process arrival and capability of computational elements speed of execution. In this paper, a two variant of static and dynamic algorithms are reviewed for various cloud based multicore environments. The organization of the paper is given as follows: Section 2 describes the literature review and section 3 explains the performance metrics associated with the load balancing algorithms. Finally section 4 concludes the paper. 2. LITERATURE REVIEW In this method, the performance (Motwani & Raghavan [1], Zhong Xu & Rong Huang [2] of all the processing elements are calculated before starting any task. This distributed multicore system follows the strategy of master slave processing. The master node allocates the task to various nodes based on the estimated load and arrival time of the tasks to the ready queue. Sometimes the slave nodes will be uded to calculate the work load of the incoming requests. In the McEntire et al [3] algorithm, a designated computational element selects a site for the execution of the new task. Whenever the task is created, a lightly loaded processing element is chosen depending on the complete picture of workload. Depending on the tasteful information, the load balancing algorithm quicklydecides upon the loadallocation and migration. Unlike the static kind of algorithms, the dynamic algorithms proposed by Malik[4], Wang & Morris[5] allocates the tasks dynamically when one of the processors becomes under- loaded. In [6], the routing for hybrid wireless network becomes complicated with respect high transmission rate. When the intermediate hops are preoccupied with data then the load balancing algorithm find out a different route based on the estimated data load during data transfer. The load balancing [7] forcloud willbeextremelycomplexasthereare multiple nodes requested at the same time for client data access. A novel methodology related to the workload distribution is implemented on the RAID systems for the better throughput and performance. IOT systems[8] have less memory and less power active devices wherein the real challenge is to find out the load of data transfer in web based IOT application. Data transfer to the web servers are implemented with the help oftraditional sender oriented load balancing algorithms. Mohammed et al.,[9] proposed a server consolidation methodology for
- 2. International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056 Volume: 09 Issue: 06 | June 2022 www.irjet.net p-ISSN: 2395-0072 © 2022, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page 2536 incorporating load balancing algorithms in cloud based environments. Cloud based datacenters are taken into consideration with master slave activation and the master nodeor host is responsible for data transferforbalancingthe load in the distributed environment. The overall survey [10] of load balancing algorithm along with the different performance metric comparison has been depicted in the work of Kashyap et al.,Mainly the distributed file system architecture is evaluated for the advanced cloud based systems. Central queue algorithm proposed by William Leinberger [11] based on the working of simple FIFO mechanism.Wheneverthereisresourcerequestrequirement comes to the centralized node which has the queue, it will be inserted at the beginning of the queue and whenever the processor becomes available then the request will be removed and given for execution. Similarly, in-order to implement the load balancing concept, the task migration can be done in FIFO queue order. In homogeneous system [11], the real challenge predominantly lies in the allocation of requests to the respective servers where the data is present. Resources are placed in aresource pool and the resource manager is responsible for handling the incoming requests. Parallel computing system [12] handlesload balancing with the help of master nodesandthe slave nodes will be involved only with the computational work.Tremendousamountofmassiveparallelismisachieved with the help of this novel idea. 3. EVALUATION PAPRAMETERS The performance of various load balancing algorithms is measured by the following parameters. Eliminating overload Whenever there is a overload situation which cannot be handled by the existing algorithms then the load balancing algorithmsstopsexecutionandeliminatesfurtheracceptance of workload from the clients. Resource availability This parameter will be very crucial for the user level satisfaction as the user always wantsa seamless executionof tasks for their different requests. Availability always related to the fault tolerance of the distributed multicore systems. Fault tolerance makes the system work even in the case of failure of the system. So for better performance of workload migration fault tolerance is very important. Accuracy level Identification Static load balancing algorithms provide more accuracywith respect to the identification of workloadasitispredefined.In the dynamic algorithms the speculation of work load sometimes depends on the delays inculcated on the network and leads to incorrect prediction of load. Single Point of Access Vs Distributed In the single point load balancing algorithms every load balancing decision as well as task migration will be happening at only one centralized node and the rest of the nodes or sites will be actively participating in the computational work. Ideally in thedistributedloadbalancing algorithm all the nodes will be involved in the load balancing decision making which makes the entiresystemafairkindog balancing the work load. Dependent and Independent Processing The decision of load migration depends on the type of execution carried on the processing elements on the shared system. If a task is divided into several subtasks and each individual subtasks are carried on different geographically distributedsystemsthenthemigrationofworkloadshouldbe carefully done by the load balancing algorithm for the final completion of the original task. For independent tasks, there will not be any complication as they can independently complete the execution. Hardware and Software Utilization Resource utilization request from the clientcan be hardware or software oriented. If more number of utilization comes from the same client and if other client requests are not handled then more adaptable dynamic strategies can be implemented. 4. CONCLUSION Basically load balancing algorithms are categorizedeitheras a static algorithm or dynamic algorithm. In this paper, we reviewed mostly static load balancing algorithms where it allocates the work to the processing elements probabilistically or deterministically and do not considerthe application level runtime scenarios. It is concluded that it is not possible to find out the time of arrival of the process and also speculating the future execution times of different workloads. We also investigated very few dynamic algorithms in which the load can be predictedwithrespectto the network condition and processing elements capabilities. In modern technology implemented on top of the distributed multicore systems, a more reliable dynamic algorithms can be incorporated for better performance. REFERENCES [1] Motwani, R & Raghavan, P 1996, ‘Randomized algorithms’, ACM Computing Surveys (CSUR), vol.28(1), pp.33-37.
- 3. International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056 Volume: 09 Issue: 06 | June 2022 www.irjet.net p-ISSN: 2395-0072 © 2022, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page 2537 [2] Zhong Xu & Rong Huang 2006, ‘Performance Study of Load Balancing Algorithms in Distributed Web Server System’, CS213 Parallel and Distributed Processing Project Report. McEntire Munro D,S & Brown, A,L 2000, ‘Evaluating partition selection policies using the CMOS garbage collector’, POSG ninth International workshop on Persistent Objects Systems, pp. 104-115. [3] Malik, S 2000, ‘Dynamic Load Balancing in a Network of Workstation’, 95.515 Research Report. [4] Wang, Z & O’Boyle, MF 2009, ‘Mapping parallelism to multi-cores: a machinelearningbasedapproach’Proc.of the 14th ACM SIGPLAN Symposium on Principles and Practice of Parallel Programming. [5] Yan Yan, Qiang Hu, Douglas M. Blough, "Load-Balanced Routing for Hybrid Fiber/WirelessBackhaul Networks", 2021 IEEE Global Communications Conference (GLOBECOM), pp.1-6, 2021. [6] N. S. Raghava and Singh Deepti, "Comparative Study on Load Balancing Techniques in Cloud Computing", Open journal of mobile computing and cloud computing, vol. 1, no. 1, pp. 18-25, August 2014. [7] Sagar Vats, Yachana Kumari, Tushar Singh, Sunil Chandra, Sushma Shirke, "LoadBalancinginSpecificIOT Systems", 2021 International Conference on Intelligent Technologies (CONIT), pp.1-6, 2021. [8] Mohammed Ala’anzy, Mohamed Othman, "Load Balancing and Server Consolidation inCloudComputing Environments: A Meta-Study", IEEE Access, vol.7, pp.141868-141887, 2019. [9] Kashyap Dharmesh and Viradiya Jaydeep, "A Survey Of Various Load Balancing Algorithms in Cloud Computing", International Journal Of Scientific & Technology Research, vol. 3, no. 11, pp. 115-119, November 2014. [10] WilliamLeinberger,George Karypis&Vipin Kumar2000, ‘Load Balancing Across Near-Homogeneous Multi- Resource Servers’, 0-7695-0556-2/00, IEEE. [11] Archer C.J., Mullins T.J, Sidelnik A. and Smith B.E. “Parallel Computing System Using Coordinator and Master Nodes for Load Balancing and Distributing Work”. United State Patent, 2010.