A survey on the performance of job scheduling in workflow application

International Journal of Computer and Technology (IJCET), ISSN 0976 – 6367(Print),
International Journal of Computer Engineering Engineering
ISSN 0976 – 6375(Online) Volume 1, Number 2, Sept – Oct (2010), © IAEME
and Technology (IJCET), ISSN 0976 – 6367(Print)
ISSN 0976 – 6375(Online) Volume 1 IJCET
Number 2, Sept - Oct (2010), pp. 69-84 ©IAEME
© IAEME, http://www.iaeme.com/ijcet.html

A SURVEY ON THE PERFORMANCE OF JOB
SCHEDULING IN WORKFLOW APPLICATION ON GRID
PLATFORMS
D.Daniel
PG Scholar
Department of Computer Science and Engineering
Karunya University, Coimbatore
E-Mail: daniel_joen@yahoo.com

Mrs. Margret Johnson
Asst.Professor
E-Mail: margret_cse@karunya.edu

D.Asir
PG Scholar
E-Mail: asir.info87@gmail.com

A.Catherine Esther Karunya
PG Scholar
E-Mail: catherineesther@karunya.du.in

ABSTRACT
Performance is the main concern of scheduling the performance driven grid
applications. The job scheduling algorithms provides the best throughput in completion
of the task or it mainly concentrates on the resource utilization. The performance is
calculated based on any one of the above mentioned objective, both for static and
dynamic scheduling. The Adaptive Dual Objective Optimization scheduling (ADOS)
algorithms takes two objectives, resource usage and completion time of a task for
measuring the performance of the schedule in workflow applications. The study shows

69

International Journal of Computer Engineering and Technology (IJCET), ISSN 0976 – 6367(Print),

ADOS algorithm delivers promising performance in three aspects: completion time,
resource usage, and robustness to resource-performance fluctuations.
Index Terms Tasks, Makespan, Resource utilization, Resource usage.
I. INTRODUCTION
To produce an effective scheduling by accounting completion time and resource
usage for workflow applications on grid platform. The proposed algorithm performs the
scheduling by accounting for both completion time and resource usage-dual objectives.
This algorithm incorporates rescheduling to deal with unforeseen performance
fluctuations effectively. The proposed algorithm delivers performance and robustness to
resource performance fluctuations.[1]
To reduce the makespan, the resource are made redundant, hence the performance
of the resource management decreases. Since the performance of the grid resource
changes dynamically, its accurate estimation of their performance is very difficult.
The majority of the grid applications fall into the interdependent task model.
These classes of applications are generally known as workflow applications. Each
composed of a number of interrelated jobs. Like traditional parallel programs, a workflow
application can be represented by a directed acyclic graph (DAG), where the nodes
represent individual jobs, and edges represent dependencies between these jobs.
Workflow applications can take advantage of a grid computing platform;
however, inter job dependencies in such applications, besides resource heterogeneity and
Dynamism, impose a great burden on scheduling. In some systems, this workflow
scheduling is left for manual dispatch by users, while other systems employ automated
workflow management platforms (WMPs) such as, Grid Flow [2]. These WMPs tend to
focus on the minimization of the application’s completion time. However, there are other
important performance considerations of WMPs, such as resource usage, load balancing,
and fault tolerance. Although some WMPs have facilities to deal with these
considerations, they often lack the capability of explicit resource usage control. Rather,
for the sake of fault tolerance, resources are overly used (task duplication).

70


II. AN ADAPTIVE RESCHEDULING STRATEGY FOR GRID
WORKFLOW APPLICATIONS.
Adaptive Rescheduling approach that can both continuously improve performance
by considering the new resources and minimize the impact caused by unexpected
resource downgrade or unavailability.
System Architecture: This system design which adapts the Planner to
dynamic grid environment via collaboration with the Executor. The system consists of
two main components Planner and Executor. The GRID Services on top of which the
Executor is built, is a collection of essential Services for any grid system. The capabilities
of each system components are defined as below:
Planner: The Planner has a collective set of subcomponents, including
Scheduler, Performance History Repository and Predictor. For each workflow application
represented as a DAG, the Planner instantiates a Scheduler instance. Based on the
performance history and resource availability, the Scheduler inquires the Predictor to
estimate the communication and computation cost with the given resource set. It then
decides on resource mapping, with the goal of achieving optimal performance for entire
workflow, and submits the schedule to the Executor. During the execution, the Scheduler
instance listens to the pre-defined events of interest, for example, addition of new
resources and significant variance of job performance evaluates the event and reschedules
the application if necessary. In the mean time, the Scheduler updates the Performance
History Repository with the latest job performance information to improve the estimation
accuracy subsequently. As each instance of Scheduler manages an individual DAG, it
will focus on the events specifically associated with the DAG it manages to continuously
improve the makespan. By collectively working with other instances of Scheduler, this
multiple-instance scheme can further improve the overall system performance [7].
Executor: The Executor is an enactment environment for workflow
applications, can be further decomposed into Execution Manager, Resource Manager and
Performance Monitor according to their respective roles in the run time environment. The
Executor supports advance reservation of resources. The Execution Manager receives the
DAG and executes it as scheduled. It is also responsible for getting job input file ready

71


and executing the job on mapped resource. Upon arrival of a schedule, the Resource
Manager will reserve the resource as per the schedule. If the arriving schedule is a result
of rescheduling, it revokes resource reservation for replaced schedule before making new
reservations. As part of the collaboration, the Resource Manager and Performance
Monitor update the Planner with information and event subscribed by the Planner.
A. HEFT – Based Adaptive Rescheduling: AHEFT
Adaptive scheduling strategy, which is an HEFT-based adaptive rescheduling
algorithm, referred to as AHEFT hereafter. Specifically, we use HEFT heuristic to
implement the schedule (S0, P,H) method in the generic algorithm. For consistence
purposes, A workflow application is represented by a direct acyclic graph, G = (V, E),
where V is the set of v jobs (nodes) and E is the set of edges between jobs. Each edge (i,
j) 2 E represents the precedence constraint such that job ni should complete its execution
before job nj starts. Data is a v × v matrix of communication data, where data i,k is the
amount of data required to be transmitted from job ni to job nk. R is a set of resources
which represent computation units.

Figure 2.1 The diagram of the system design.

72


The variable clock is used as logical clock to measure the time span of DAG
execution, it is initially set as 0 when the DAG starts to execute. When the DAG finishes
successfully, the clock reads as the Makespan of the DAG. A job cannot start without all
required inputs available on the resource on which the job is to execute. Such inputs are
in turn the outputs from immediate predecessor jobs. the upward rank is calculated for
each job and sorted in non increasing order, which corresponds to significance order how
the individual job affects the final Makespan[7][9]. The basic concept of this algorithm is
to select the “best” resource which minimizes the earliest finish time of the job currently
with highest upward rank and remove the job from unscheduled job list once it is
assigned with resource. The resource selection process repeats until the list is empty.
III. GRIDFLOW: WORKFLOW MANAGEMENT FOR GRID
COMPUTING.
Grid computing is becoming a mainstream technology for large-scale distributed
resource sharing and system integration. Workflow management is emerging as one of
the most important grid services. A workflow management system for grid computing
called Grid Flow. The main functionalities of grid workflow management include
workflow construction, simulation, scheduling, execution, monitoring, conflict solving,
and so on. A grid environment presents a number of new challenges for the workflow
management system:
CROSS – DOMAIN:
The process of a grid workflow encompasses multiple administrative domains
(organizations). The lack of central ownership and control results in incomplete
information and many other uncertain factors.
Dynamism: Since grid resources are not entirely dedicated to the environment,
computational and networking capabilities can vary significantly over time. Application
performance prediction becomes difficult and real-time resource information update
within a large-scale global grid becomes impossible [2].
A. GRID WORKFLOW MANAGEMENT.
This work focuses more on service-level support such as grid workflow
management and scheduling, a Grid Flow user portal is also developed that provides a

73


graphical user interface (GUI) to facilitate the composition of grid workflow elements
and the access to additional grid services. The system is designed so that workflow
management follows the same layered framework as that of resource management,
including global grid workflow management and local grid sub workflow scheduling.
The implementation of grid workflow management is carried out at multiple layers:
Task: Tasks are the smallest elements in a grid workflow. Task scheduling is
implemented using Titan, and as stated, this work focuses more on the sub-workflow and
workflow levels of management and scheduling.
Sub-workflow: A sub-workflow is a flow of closely related tasks that is to be
executed in a predefined sequence on grid resources of a local grid (within one
organization). Conflicts occur when tasks from different sub-workflows require the same
resource simultaneously.
Workflow: A grid application can be represented as a flow of several different
activities, each activity represented by a sub-workflow. These activities are loosely
coupled and may require multi-sited grid resources. Simulation, execution and
monitoring services can be provided [2].
1) GRIDFLOW USER PORTAL
The Grid Flow portal is an integrated environment that enables users to construct
a grid workflow and access grid services. To construct a grid workflow, a user needs to
define properties of each sub-workflow and task and their execution sequences. In
general, a sub-workflow or a task can have several pre- and post- activities the portal also
provides direct user interfaces to the information and performance services. However, if
the user does not know anything about the available grid services and resources, he can
submit the workflow to the global workflow management system, which will provide the
services automatically.
2) GLOBAL GRID WORKFLOW MANAGEMENT
The global grid workflow management system receives requests from the Grid
Flow portal with XML specifications of grid workflows, and provides three main
functionalities:

74


Simulation: Simulation takes place before a grid workflow is actually executed,
during which time a workflow schedule is achieved. The simulation results can be
returned to the Grid Flow portal for user agreement or passed directly to the execution
engine.
Execution: A grid workflow is executed according to the simulated schedule. Due to
the dynamic nature of the grid environment, the schedule may not be executed
accordingly. When large delays of some sub workflows occur, the rest or whole of the
workflow may be sent back to the simulation engine and rescheduled.
Monitoring: Global grid workflow management also provides interfaces that provide
access to real-time status reports of task or sub-workflow execution. GGWM
algorithm is used [11].
3) LOCAL GRID SUB-WORKFLOW SCHEDULING
Scheduling a flow of tasks onto grid resources within a local grid is very similar
to the process that schedules a workflow onto different local grids introduced above. One
Important difference is that the local grid sub workflow scheduling has to deal with
multiple tasks that may belong to different sub-workflows. The execution time has to be
estimated with the extra consideration of conflicts, which may occur when multiple tasks
require the same grid resource at the same time. LGSS algorithm is used.
IV. PERFORMANCE-EFFECTIVE AND LOW-COMPLEXITY
TASK SCHEDULING FOR HETEROGENEOUS COMPUTING.
The application scheduling problem has been studied and various algorithms have
been proposed mainly for the homogeneous processors. There are only few algorithms
for heterogeneous processors, but they are high scheduling cost and they do not deliver
quality schedule with lower cost. Two new algorithms have been proposed for
heterogeneous processors which provide fast scheduling time and high performance,
HEFT and CPOP algorithm.
A. HETEROGENEOUS – EARLIEST – FIRST – TIME (HEFT)
ALGORITHM
The HEFT algorithm is an application scheduling algorithm for bounded number
of heterogeneous processors, which has two major phases a task prioritizing phase for

75


computing the priorities of all tasks and a processor selection phase for selecting the tasks
in order of their priorities and scheduling each selected task on its “best” processor,
which minimizes the task’s finish time.
Task prioritizing phase: This phase requires the priority of each task to be
set with upward rank value, which is based on the mean computing and mean
communication cost. The task list is generated by sorting the task by decreasing order of
rank value. Tie breaking is done randomly. There can be alternative policies for tie
breaking, such as selecting the task whose immediate successor tasks has higher upward
ranks. Since these alternate policies increase the time complexity, random selection
strategy is preferred.
Processor selection phase: For most of the task scheduling algorithm, the
earliest available time of a processor pj for a task execution is the time when pj completes
the execution of its last assigned task. However, the HEFT algorithm has an insertion-
based policy which considers the possible insertion of a task in an earliest idle time slot
between two already – scheduled tasks on a processor. The length of an idle timeslot, i.e.
the difference between execution start time and finish time of two tasks that were
consecutively scheduled on the same processor, should be at least capable of computation
cost of the task to be scheduled.
B. CRITICAL PATH-ON-A PROCESSOR (CPOP) ALGORITHM
CPOP algorithm has two phase, task prioritizing and processor selection phase. It
uses a different attribute for setting the task priorities and a different strategy for
determining the best processor for the selected task.
Task prioritizing phase: Upward rank and downward rank for all tasks are
computed using mean computation and mean communication cost. CPOP uses critical
path for the given application graph. The length of the critical path is the sum of
computation cost of tasks on the path and intertask communication cost along the path,
critical path length is equal to the entry task’s priority. Initially the entry task is selected
and marked as critical path task; an immediate successor of the selected task that has the
highest priority value is selected and marked as critical path task. This process is repeated

76


until the exit node is reached. For tie break, successor which has highest priority is
selected.
Processor selection phase: Critical path processor is the one that minimize
the computation cost of the tasks on critical path. If the selected task is on critical path,
then it is scheduled on the critical path processor. Otherwise, assigned to processor which
minimize the earliest execution finish time of task. In both cases insertion policy is
consider.
V. A TASK DUPLICATION BASED BOTTOM-UP ALGORITHM
FOR HETEROGENEOUS ENVIRONMENTS
DBUS algorithm traverses the DAG in a bottom-up fashion while taking
advantage of task duplication and task insertion. The makespans generated by the DBUS
algorithm are much better than those generated by the existing algorithms, HEFT,
HCPFD and HCNF.
A. DBUS: Duplication-based Bottom-Up Scheduling Algorithm
The DBUS algorithm consists of a critical path-based listing phase followed by
duplication based scheduling phase. In the traditional approach of top-down DAG
traversal, duplication is carried out when a task's finish time can be improved with
duplication of its critical parent on the same processor with the task itself. This means
that a task with multiple children can be duplicated on multiple processors if its
duplication helps improving its children's finish time on those processors [6]. The
drawback in this approach is that, the duplication of the parent task on each of such
processors is done independently. Therefore the positions of the parent task on relevant
processors are decided without any optimization with respect to each other. By traversing
the DAG in a bottom up fashion, the proposed algorithm first schedules all children of a
task before scheduling the task on as many processors as necessary. Consequently, it is
more likely to make good duplication decisions since all copies of the parent task are
considered at the same time.
In addition, the stop criterion for duplication is not determined by the number of
duplications already carried out, but by the quality of the current schedule. Since the
number of beneficial duplications may differ significantly across different problems,

77


DBUS offers superior performance compared to algorithms that limit duplication by the
number of duplicates. Furthermore, the DBUS algorithm is an insertion based algorithm
that allows tasks to be scheduled at the first available time slot that can accommodate
themselves. Task insertion based algorithms have better chances of finding shorter
schedule lengths compared to non-insertion based ones. There are two phases: listing
phase and scheduling phase [5].
Algorithm Task insertion Restrictions on duplication
HEFT Yes No duplication
HCPFD No Only critical parent of a task
HCNF No Only critical parent of a task
DBUS Yes No restriction
Table 5.1 Comparison of Scheduling Algorithms
VI. ON THE PERFORMANCE OF A DUAL OBJECTIVE
OPTIMIZATION MODEL FOR WORKFLOW APPLICATIONS ON
GRID PLATFORMS
The primary objective of scheduling the workflow application is to minimize the
completion time of the application. However, they tend to achieve this objective at the
expense of redundant resource usage. The proposed algorithm performs the scheduling by
accounting for both completion time and resource usage - dual objectives. Since the
performance of grid resources changes dynamically and the accurate estimation of their
performance is very difficult, this algorithm incorporates rescheduling to deal with
Unforeseen performance fluctuations effectively.
WORKFLOW APPLICATION
The majority of applications in the grid environment fall into interdependent task
model known as Workflow applications. Each workflow application is composed of
number of interrelated jobs. Like traditional parallel programs, a workflow application
can be represented by a directed acyclic graph (DAG), where the nodes represent
individual jobs, and edges represent dependencies between these jobs.
Workflow applications can take advantage of a grid computing platform;
however, interjob dependencies in such applications, besides resource heterogeneity and
dynamism, impose a great burden on scheduling. In some systems, this workflow
scheduling is left for manual dispatch by users, while other systems employ automated

78


workflow management platforms (WMPs) [1].These WMPs tend to focus on the
minimization of the application’s completion time. However, there are other important
performance considerations of WMPs, such as resource usage, load balancing, and fault
tolerance. Although some WMPs have facilities to deal with these considerations, they
often lack the capability of explicit resource usage control. Rather, for the sake of fault
tolerance, resources are overly used (task duplication)[8].
ADAPTIVE DUAL OBJECTIVE SCHEDULING (ADOS)
ALGORITHM
This paper address the problems of scheduling workflow applications in grid and
propose an Adaptive Dual Objective Scheduling (ADOS) Algorithm as a novel semi
dynamic scheduling heuristic, which statically generates the initial schedule using an
evolutionary technique and adapts it dynamically as the performance of resources
changes; hence, the algorithm is semi dynamic. The two main strengths of ADOS are 1)
the good quality of the output schedules with small resource usage and 2) the adaptability
to performance fluctuations on resources. The first strength is achieved by iteratively
improving an initial random schedule using a branch-and-bound technique and genetic
operator (mutation) improvements are made not only in completion time but also in
resource usage. Resource usage is equally seriously taken into account compared to
completion time, because the overuse of grid resources shared by many users, so at any
given time, there might be a number of jobs running on those resources degrades the
overall performance of the grid. The second strength is made possible by incorporating a
systematic rescheduling strategy. Specifically, ADOS initiates a rescheduling event if a
job finishes later than expected and its late completion results in an increase in the overall
application completion time. Rescheduling occurs with all of the remaining jobs that are
not yet running [1][10].
While the task scheduling problem in heterogeneous computing systems with
perfectly accurate performance information on resources and applications still remains
very difficult (NP-hard), uncertainties on resource performance and the lack of control
over grid resources make workflow scheduling even more challenging. This algorithm for

79


workflow scheduling with the detailed description of our specific approaches to those
issues in such a scheduling problem. Unlike many other workflow scheduling schemes,
we consider both makespan and resource usage to be equally important and take this into
account in our scheduling model. Efficient resource usage is crucial in grid scheduling
because 1) a grid consists of multiple sites administered by different entities that use their
own resources for other tasks beside the grid jobs and 2) due to the fluctuations and
uncertainty surrounding sites in a grid system, lower resource usage not necessarily the
minimization of the number of resources used, rather the minimization of resource time
means lower overall variance in the expected completion time (makespan) of an
application. Rescheduling is another technique adopted to increase the practicality of
ADOS.
A. WORKING OF ADOS
To start with, ADOS generates a random schedule, this initial solution undergoes
the manipulation process of ADOS repeatedly for further improvement in makespan and
resource usage. This schedule manipulation involves a branch-and-bound-style technique
and two types of mutation (point and swap). Each job in the initial random schedule is
tried on each available host to check whether any of these new matches shortens the
current makespan. If one or more matches better than the original match are identified,
the host on which the makespan is reduced the most is selected. ADOS keep track of the
two best schedules (relaxed and strict) possibly the same schedule based on makespan
and on makespan and resource usage, respectively. At first glance, the reason for
maintaining the best schedule based only on makespan, which is not clear, ADOS
determines the quality of schedule based on both makespan and resource usage. We have
observed that the application of such strict criteria throughout the entire scheduling
process tends to produce very good schedules in terms of resource usage; however, this
strict schedule selection scheme raises the issue that schedules are less attractive, as far as
makespan is concerned. This observation has led to the incorporation of a loosened
schedule selection scheme as an effective solution to the issue. This scheme helps shorten
the makespan noticeably, as the result of a compromise between makespan and resource
usage. The scheme implicitly takes into account resource usage. Specifically, the resource

80


usage of the best schedule is guarded to remain at a reasonable level by , which can be
seen as the quality guarantor of ADOS for resource usage. Note that S* is a slight
modification of S’. Thus, the degree of sacrifice made to the resource usage of S* is
minimal.
At the end of each iteration, mutation is considered if no improvement is made
during the current iteration. ADOS randomly choose a mutation method between point
and swap mutations and mutates each job to generate substantially different schedules.
Note that mutation only occurs with unscheduled jobs when rescheduling. The mutated
schedule is then used as the current schedule and the strict best schedule in the next
iteration. If there have been some improvements in the current iteration, S is passed onto
the next iteration for further improvements. This is because changes made to jobs with
low b-level values (low-priority jobs) may enable better job-host matches, in the next
iteration, for unchanged high-priority jobs leading to improvement in the quality of the
current schedule as well as the best schedules . This schedule manipulation process
repeats for a predefined number of iterations.
Now, jobs in the best schedule are dispatched to their assigned hosts as they
become ready, i.e., their predecessor jobs have finished. During this actual job dispatch
process, there might be cases in which some jobs are delayed for unacceptably long to
complete their execution, these trigger rescheduling events. Here, it is very crucial how to
define “unacceptably long.” For each job, ADOS computes the actual latest finish time,
which is a modification of LFT. The modification is made because ADOS uses job
insertion. The ALFT of a job is an indicator of whether the delay in the completion of the
job is acceptable. In other words, the late completion of a job does not affect the
makespan of a given workflow application as long as the time of the completion is no
later than the actual latest finish time of the job. Hence, the delay is acceptable [1].

81


Figure 6.1 Structure of Ados algorithm
VII. EVALUATION AND COMPARISON
Adaptive scheduling strategy based HEFT-based adaptive rescheduling algorithm
(AHEFT) has the advantage of continuously improves the performance, considering the
new resource and minimizes the impact caused by unexpected resource down grade or
unavailability [7]. Drawbacks of adaptive rescheduling technique are it takes time in
rescheduling the jobs and to implement the collaboration model, rescheduling has to be
integrated with advance resource reservation and resource availability prediction model.
The grid flow gives the Advantage of grid performance service comprises performance
prediction capability with a new application response measurement technique, which can
be used to enable prediction-based scheduling as well as response-based scheduling. But
the disadvantage such as the process of a grid workflow encompasses multiple
administrative domains (organisations)[7]. The lack of central ownership and control
results in incomplete information and Computational and networking capabilities can
vary significantly over time in the grid environment. Application performance prediction

82


becomes difficult and real-time resource information update within a large-scale global
grid becomes impossible, which lower its performance.
Critical Path-on-a processor (CPOP) Algorithm and Heterogeneous Earliest First
Time (HEFT) Algorithm, both gives more or less same performance measures as high
performance and fast schedule time. But has slight disadvantage of high schedule cost.
Duplication-based Bottom-Up Scheduling Algorithm (DBUS) gives uses both task
insertion and task duplication. It also gives the facility to minimize the schedule length
[6]. It does not impose any restriction on number of task duplication and Task duplication
mainly causes an increase of resource usage which causes disadvantage to the algorithm.
The ADOS algorithm has the highest possibility for reducing the makespan of the task,
which is the total amount of time required from the start of the first job to the end of the
last job. It also reduces the resource usage .the another advantage of this algorithm is
semidynamic, which could adapt to the dynamic changes that happens in the grid
environment.
VIII. CONCLUSION
In this paper we surveyed four scheduling algorithms. Adaptive scheduling
strategy based HEFT-based adaptive rescheduling algorithm (AHEFT) which is follows
dynamic scheduling. Critical Path-on-a processor (CPOP) Algorithm and Heterogeneous
Earliest First Time (HEFT) Algorithm, Duplication-based Bottom-Up Scheduling
Algorithm (DBUS) which follows static scheduling. ADOS algorithm follows
Semidynamic scheduling strategy, which schedules the jobs statically and adapts its
schedule as per the dynamic changes occur in the grid environment using rescheduling.
This algorithm also provides dual object and improves the job scheduling.
REFERENCE
[1] Young Choon Lee, Member, IEEE, Riky Subrata, and Albert Y. Zomaya, Fellow,
IEEE “On the Performance of a Dual-Objective Optimization Model for Workflow
Applications on Grid Platforms,” Proc, IEEE Transactions On Parallel And
Distributed systems, Vol. 20, N0. 9, September 2009.

83


[2] J. Cao, S.A. Jarvis, S. Saini, and G.R. Nudd, “GridFlow: Workflow Management for
Grid Computing,” Proc. Third IEEE/ACM Int’l Symp. Cluster Computing and the
Grid (CCGrid ’03), pp. 198-205, 2003.
[3] H. Casanova, “Simgrid: A Toolkit for the Simulation of Application Scheduling,”
Proc. First IEEE/ACM Int’l Symp. Cluster Computing and the Grid (CCGrid ’01), pp.
430-437, 2001.
[4] R. Wolski, “Dynamically Forecasting Network Performance Using the Network
Weather Service,” Proc. Sixth IEEE Int’l Symp. High Performance Distributed
Computing (HPDC ’97), pp. 316-325, 1997.
[5] H. Topcuoglu, S. Hariri, and M. Wu, “Performance-Effective and Low- Complexity
Task Scheduling for Heterogeneous Computing,” IEEE Trans. Parallel and
Distributed Systems, vol. 13, no. 3, pp. 260-274, Mar. 2002.
[6] D. Bozdag, U. Catalyurek, and F. Ozguner, “A Task Duplication Based Bottom-Up
Scheduling Algorithm for Heterogeneous Environments,” Proc. 19th Int’l Parallel and
Distributed Processing Symp. (IPDPS ’05), Apr. 2005.
[7] Z.Yu and W.Shi, “An Adaptive Rescheduling Strategy for Grid Workflow
Applications,” Proc. 21st Int’l Parallel and Distributed Processing Symp. (IPDPS),
2007.
[8] Y. Gil, V. Ratnakar, E. Deelman, G. Mehta, and J. Kim, “Wings for Pegasus:
Creating Large-Scale Scientific Applications Using Semantic Representations of
Computational Workflows,” Proc.19th Conf. Innovative Applications of Artificial
Intelligence (IAAI ’07), pp. 1767- 1774, 2007.
[9] M. Wieczorek, R. Prodan, and T. Fahringer, “Scheduling of Scientific Workflows in
the ASKALON Grid Environment,” ACMSIGMOD Record, vol. 34, no. 3, pp. 56-62,
Sept. 2005.
[10] G. Singh, E. Deelman, G. Mehta, K. Vahi, M.-H. Su, G.B. Berriman,J. Good, J.C.
Jacob, D.S. Katz, A. Lazzarini, K. Blackburn, andS. Koranda, “The Pegasus Portal:
Web Based Grid Computing,”Proc. 20th Ann. ACM Symp. Applied Computing
(SAC ’05),pp. 680-686, 2005.

84

A survey on the performance of job scheduling in workflow application

More Related Content

What's hot (13)

Similar to A survey on the performance of job scheduling in workflow application (20)

More from iaemedu (20)

A survey on the performance of job scheduling in workflow application