Applying Advance Reservation to Increase Predictability of Workflow Execution
on the Grid
Marek Wieczorek, Mumtaz Siddiqui, Alex Villaz´ on, Radu Prodan, Thomas Fahringer
Institute of Computer Science, University of Innsbruck
Technikerstraße 21a, A-6020 Innsbruck, Austria
In this paperwe presentan extensionto devise and imple-
ment advance reservation as part of the scheduling and re-
source management services of the ASKALON Grid appli-
cationdevelopmentand runtimeenvironment. The schedul-
manager about resources capable of processing tasks in the
shortest possible time. We introduce progressive reserva-
tion approach which tries to allocate resources based on a
fair-share principle. Experiments are shown that demon-
strate the effectiveness of our approach, and that reflect dif-
ferent QoS parameters including performance, predictabil-
ity, resource usage and resource fairness.
Scheduling and resource management play a key role for
effective execution of applications on the Grid. A sched-
uler is used to make a plan for the execution, determining
which tasks should be executed on which resources and in
which order. Resource management is responsible for ser-
vices such as registration and provision of resources. The
scheduler optimizes a goal function, usually execution time
or economic cost. Created in this way schedules are subse-
Highly dynamic and unreliable Grid environments make
any assumptions concerning resource availability and exe-
able schedule and proper reservationof resources constitute
one of the largest challenges in Grid computing. Grid envi-
ronments cannot usually guarantee that execution requests
will be fulfilled within expected time intervals. Moreover,
requests coming from other users can always disrupt al-
ready taken scheduling decisions. Time-critical applica-
tions, which are an important class of Grid applications,
expected execution time. Advance reservation of resources
provides a possible solution to this problem. A user can re-
serve resources, in order to be sure that within a requested
period of time resources will be available.
This paper investigates the impact of resource reserva-
tions on different aspects of application execution, repre-
sented by different criteria comprising execution time, pre-
dictability, resource usage and fairness. Predictability can
beconsideredas the mostimportantcriterion,becauseit has
a substantial impact on the execution of time-constrained
applications. We have chosen scientific grid workflow ap-
plications  as the class of computational applications to
validate our approach. Workflow applications consist of ex-
The rest of the paper is organized as follows. In the next
section we describe briefly the related work on advance
reservations for the Grid. Later, we present the architecture
of our Grid environment, focusing on scheduling and re-
source management. Section 5 and 6 provide details of our
reservation and scheduling services respectively. Section
7 describes the applied experimental methodology, and the
methodsused to evaluatethe results. Results and discussion
are presented in the Section 8. The last section concludes
the paper and discusses future work.
3 Related work
The Maui  scheduler is an advanced job scheduler
for cluster systems in which an advance reservation scheme
enables future allocation of local resources. On a Grid site,
Managers (LRM), such as PBS , SGE , LSF .
Maui does not provide negotiation mechanism at resource
and Grid level, and requiresthe Grid job submissionservice
to be modified in order to be reservation aware.
GARA  provides an interesting proposal for an ar-
chitecture supporting advance reservation and co-allocation
of resources in large Grid environments incorporating re-
ered by this work, nordoes the system provideany practical
solution to the problem of fairsharing.
ThomasRoebliz et al. presentsan elastic Grid reservation
with user defined optimization policies and co-reservation
with a concept of virtual resources . No negotiation or
up to 50%. The mixedstrategy presents a goodfair-sharing,
which is about 2 times better than for strategies with reser-
Table 2. Execution without reservations - odd
numbers, and with progressive reservations
with short reservation requests (20% longer
than the predictions) - even numbers
9Conclusions and Future Work
This paper presents a scheduling and an advance reser-
vation service for execution of scientific workflow appli-
cations on the Grid. We developed a resource manage-
tomizable negotiation mechanism. We proposed the pro-
gressive reservation approach which appears to be a good
solution for Grid environments dedicated for execution of
scientific workflows. We have implemented our approach
in the ASKALON application development and execution
environment. We demonstrated that advance reservation
can have a major impact on executiontime and can increase
considerably predictability of a Grid environment. We also
showed the importance of requesting for longer reservation
time periods, because of the low reliability of execution
time predictions. This may provide very high probability
thattheexecutionwill finishwithinthe reservedtime. How-
ever, this happens at the cost of lower resource usage and
fairness of resource distribution. Fairness can be increased
by employing a special reservation strategy, for instance
the progressive reservation strategy proposed in this paper,
which gives the best results when the Grid is not saturated.
Finally, we showed that the strategies without reservations
and with progressive reservations can be applied together,
and result in good performance and fairness.
In the future, we plan to investigate more dynamic reser-
vation algorithms for heavily loaded Grid environments.
We plan to study how to increase the resource usage and
consider network traffic as an important aspect of workflow
scheduling and execution.
 NorduGrid project. http://www.nordugrid.org/.
 WS-GRAM. http://www.globus.org/toolkits/docs/ 4.0/exe-
 G. M. Amdahl. Validity of the single processor approach to
achieving large scale computer capabilities. In AFIPS joint
computer conference, volume 30, pages 483–485, 1967.
 P. Blaha, K. Schwarz, G. Madsen, D. Kvasnicka, and
J. Luitz. WIEN2k: An Augmented Plane Wave plus Local
Orbitals Program for Calculating Crystal Properties. In-
stitute of Physical and Theoretical Chemistry, TU Vienna,
 E. Elmroth and J. Tordsson. A Grid Resource Broker Sup-
porting Advance Reservations and Benchmark-based Re-
source Selection. In State-of-the-art in Scientific Comput-
ing., volume 3732 of Lecture Notes in Computer Science,
pages 1077–1085. Springer Verlag, Oct. 2005.
 J.-J. H. et al. Scheduling precedence graphs in systems
with interprocessor communication times. J. Comput., (18
 T. Fahringer, R. Prodan, R. Duan, F. Nerieri, S. Podlip-
nig, J. Qin, M. Siddiqui, H.-L. Truong, A. Villazon, and
M. Wieczorek. ASKALON: A Grid Application Develop-
ment and Computing Environment.
Workshop on Grid Computing (Grid 2005), Seattle, USA,
Nov. 2005. IEEE Computer Society Press.
 I. Foster, C. Kesselman, C. Lee, R. Lindell, K. Nahrstedt,
and A. Roy. A Distributed Resource Management Architec-
turethat Supports Advance Reservations and Co-Allocation.
In Proceedings of the International Workshop on Quality of
Service, pages 27–36, London, UK, June 1999. IEEE.
 R. B. J. Yu. A Taxonomy of Scientific Workflow Systems
for Grid Computing. SIGMOD Record, 34(3):44–49, Sept.
 Load Sharing
 Maui Scheduler. http://www.supercluster.org/maui/.
 A. S. McGough, A. Afzal, J. Darlington, N. Furmento,
A. Mayer, and L. Young.
through Reservation and Performance Modelling. The Com-
puter Journal, 48(3):358–368, 2005.
 A. G. project. http://www.austriangrid.at.
 T. Roeblitz, F. Schintke, and Wendler. Elastic grid reser-
vations with user-defined optimization policies.
ceedings of the Workshop on on Adaptive Grid Middleware,
 M. Siddiqui, A. Villazon, and T. Fahringer. Grid capacity
planning for optimized qos with negotiation-based advance
reservation. In ACM, editor, International Conference for
HighPerformance Computing, Networkingand Storage (Su-
perComputing), SC06, Tampa, USA,November 11-172006.
 G. Singh, C. Kesselman, and E. Deelman. Technical report,
University of Southern California, 2005.
 Sun Grid Engine. http://gridengine.sunsource.net/.
 The Portable Batch System. http://www.openpbs.org.
In 6th International
Making the Grid Predictable