An Analytical Model of the Service Provisioning
Time within the Harmony Network Service Plane
A. Willner
University of Bonn, Germany
Inst. of Computer Science 4
willner@cs.uni-bonn.de
J. Ferrer Riera, J. Garcia-Espin, S. Figuerola
i2CAT Foundation, Spain
Network Technologies Cluster
{jordi.ferrer,jage,sergi.figuerola}@i2cat.net
M. De Leenheer, C. Develder
Ghent University - IBBT, Belgium
Dept. of Information Technology
{marc.deleenheer,chris.develder}@intec.ugent.be
Abstract—Grid computing aims at offering standardized access
to heterogeneous and distributed resources for scientific commu-
nities. However, in order to support emerging next generation
Grid applications with specific Quality of Service requirements,
the interconnecting networks have also been considered as first-
class allocable Grid resources and have been also taken into
account for the co-scheduling process. In the last few years,
a number of network resource provisioning systems were de-
veloped, however, without providing specific analysis on the
scalability of potential architectural design alternatives. Our
approach is to formulate a fundamental analytical model to
evaluate the expected service provisioning time using different
architectures as a function of the involved transport domains.
To validate our results, we have used measurements obtained
from the European IST-FP6 Phosphorus project testbed. The
main contribution is to provide an instrument to obtain reference
values to support architectural design decisions even in an early
stage of the development phase.
Index Terms—Network Service Plane, Grid Computing, Per-
formance Evaluation, Bandwidth on Demand
I . I N T R O D U C T I O N
Since its emergence more than a decade ago in the context
of the I-WAY [1] project, the concepts behind Grid computing
have become increasingly popular and have been widely
adopted in the field of high performance computing. This
is compounded by the popularity of cloud computing and
virtualisation of both network and IT resources Within this
scope, dynamic resource co-allocation and end-to-end dynamic
provisioned services became one of the main research areas.
The fact of considering the network as a first-class Grid
resource and integrate it in the multidomain scheduling process,
requires some developments of specific Quality of Service
(QoS) features that the underlying infrastructure has to support,
and an agreement-based resource management system.
Currently, several so called Network Service Planes (NSPs)
architectures are used to provide this capability to a higher
level. The challenge is to analyze the scalability of such an
NSP considering both job workloads and different available
network topologies. As stated in [2], performance measurement
analysis has become an important tool to decide the best
network topology and must, therefore, be carefully chosen.
Defining and executing a performance evaluation of Grid-
enabled architectures is not a straightforward task, since
the requirements considered in this manuscript are manifold:
correctness in the selection of the metrics, realistic workload
models, and accurate workload generators [3].
So far, a number of national and international projects have
focused on this particular research area of network provisioning
with Grid service awareness. The outcome of one of these
projects, the IST-FP6 Phosphorus project, was the Harmony
NSP [4]. It serves as the basis for the following discussion as
its design allows to handle different architectures.
In this paper, we assume a multidomain, multitechnology, and
multivendor scenario, where each administrative domain runs
under a local Network Resource Manager (NRM). The NSP is
populated by different entities. Depending on the relationships
and interaction patterns among them and the role of each single
entity, the NSP can operate under centralized, hierarchical,
daisy-chained, hybrid, or meshed architectures, which are also
referred to as deployment models in this article.
Based on these assumptions, we have developed an analytical
model, which describes the dependencies and communication
workflow between the involved systems. This model is focused
on the end-to-end service provisioning time and can be used to
predict the expected delays in the service provisioning process.
In order to validate our model we have compared the results
with measurements gained from an emulated Harmony testbed.
The configuration parameters for both the testbed and the model
were acquired from the actual IST-FP6 Phosphorus testbed.
The key contribution of this paper is to provide a methodol-
ogy to analyze the scalability of a chosen NSP architecture as
a function of the involved transport domains given different
deployment models. Overall, this paper helps to construct a
foundation for further network research and developments in
the field of resource co-allocation and generic network service
interfaces in heterogeneous environments.
The remainder of the paper is structured as follows. We give
a brief overview of related work in the context of dynamic,
on-demand bandwidth allocation systems in Sec. II. In Sec. III,
we state the problem addressed and the terminology used along
the paper. We also present the possible architectures and the
communication protocol used within the service plane. In the
subsequent Sec. IV the analytical model is presented. The
results obtained from performance analysis of the model are
given, compared, and discussed in Sec. V. Finally, we close
giving some conclusions and considerations and describe future
work in Sec. VI.