Sie sind auf Seite 1von 6

Business Transformation to SOA: Aspects of the Migration

and Performance and QoS Issues


Liam OBrien Paul Brebner Jon Gray
NICTA NICTA NICTA
Building A, 7 London Circuit, Building A, 7 London Circuit, Building A, 7 London Circuit,
Canberra, ACT 2601, Australia Canberra, ACT 2601, Australia Canberra, ACT 2601, Australia
+61 2 62676276 +61 2 62676277 +61 2 62676275
Liam.OBrien@nicta.com.au Paul.Brebner@nicta.com.au Jon.Gray@nicta.com.au

ABSTRACT programs to determine what strategies they will pursue in the future,
Organizations face challenges to be more adaptable and transform to what business processes will support those strategies and determine
meet new customer demands with fewer resources and streamlining what IT systems are needed. As part of this rationalization many
of its business activities. There is a growing move to introduce organizations are moving to SOA based systems. There are various
SOAs with their promise of cost-efficiency, agility, adaptability and aspects to the introduction of SOA from mining and reusing legacy
legacy leverage. However there are many aspects of transforming an components as services in an SOA-based environment to acquisition
organization to use SOA and many obstacles and issues that the and development of an SOA infrastructure. In section 2 we will
organization has to address when introducing SOAs. In this paper outline several aspects of introducing SOA into an organization and
we outline some of the major aspects of SOA introduction and focus outline some of the open issues where further research still needs to
on some of the open issues that still need to be tackled. In the be done.
discussion on the various aspects of SOA introduction we focus in One of the main concerns for organizations building SOA-based
on performance and QoS which are major pieces to get right if the systems should be whether or not those systems can meet Quality-
SOA implementation is to be successful. We outline some of the of-Service (QoS) requirements. The use of an SOA can have a
work that we are doing in this area and some problem areas where positive impact on some quality attributes and a negative impact on
further research is needed. others [14]. Some of the main issues for organizations are dealing
with security, performance, availability and interoperability of their
Categories and Subject Descriptors SOA-based systems. Determining exactly what the QoS
K.6.4 [System Management]: Quality Assunance. requirements are, building their systems to conform to these
requirements, analyzing the systems to make sure that they conform
General Terms and determining what their QoS is and managing all of this through
Management, Performance specifying and negotiating service level agreements (SLAs), can be
quite burdensome on an organization. We have been working on
building performance and QoS modeling and simulation tools and
Keywords we outline our current work in this area in Section 3 and outline
Service Oriented Architecture, SOA, Performance. Quality-of- some open issues where further research is needed.
Service, Service Level Agreements, Migration.
The paper is structured as follow. Section 2 outlines various aspects
of the migration to SOA. Section 3 outlines some of the work we are
1. INTRODUCTION doing the areas of performance and QoS and outlines some of the
As organizations adapt and transform their business process and IT
open issues that we are tackling. Section 4 concludes the paper.
business systems to meet new challenges and respond to customer
demands many of them are deciding to implement SOA-based
systems. The promise of cost-efficiency, agility, adaptability and 2. ASPECTS OF SOA MIGRATION
legacy leverage has attracted many organizations to try out SOA When an organization wants to introduce the use of SOA there are
and in may cases make substantial investments in new systems that various aspects that it needs to consider. These include:
are based on SOA. The identification and mining of services from its
In order to move to transform their business processes and IT existing/legacy systems
business system many organizations are undertaking rationalization The integration of common/shared (e.g. Whole-of-
Government (WofG) services provided by one
Permission to make digital or hard copies of all or part of this work for government agency and used by other agencies) services
personal or classroom use is granted without fee provided that copies are with existing/legacy systems
not made or distributed for profit or commercial advantage and that copies The acquisition and development of an SOA
bear this notice and the full citation on the first page. To copy otherwise, or
infrastructure (including middleware)
republish, to post on servers or to redistribute to lists, requires prior specific
permission and/or a fee. The development of services and development of
SDSOA08, May 11, 2008, Leipzig, Germany. applications from services
Copyright 2008 ACM 978-1-60558-029-6/08/05...$5.00.

35
The development of governance policies and procedures determining the scope, size, effort and cost of the integration and
(including establishing and monitoring of Service Level several activities for detailed modeling and simulation of
Agreements) performance and QoS concerns with the ability to be able to
generate alternative architecture solutions to satisfy different QoS
Analysis of architectural alternatives performance,
requirements. An organization can choose a particular architecture
scalability and other QoS requirements for SOA-based
alternative based on their specific requirements.
systems
For each of these aspects we will explore some of the current One of the main open issues in integration of services is how so size
research in each area and outline some of the work that we are doing an integration effort and what are the various parts that make up the
and outline some of the open issues that are still important in each cost. Items such as number of data elements, systems complexity,
area. service complexity, process complexity, applicable standards,
potential risks, etc all have an impact on cost [16]. Development of
2.1 Identification and Mining of Services methods and techniques that tackle determination of cost in SOA
When an organization moves to using SOA it does not want to projects (individual activities as well as the total cost) have still to
throw away its investment in its existing systems and has to deal be fully develop, tested and used across the industry.
with them. Therefore there is a need to determine what services are
needed and what components from existing system can be migrated 2.3 Development of an SOA Infrastructure
to fill these service needs. There are several activities that an
To run and manage SOA applications, an organization needs an
organization will need to undertake:
SOA infrastructure. An SOA infrastructure consists of several
Identify what services are to be used within its SOA. An pieces that support the main aspects of SOA including security,
organization can do this by developing an overall governance, management, orchestration and resourcing (e.g.
Enterprise Architecture, required business processes and virtualization). Gartner lists the following as major technical errors
identifying services that are needed (top-down) or it can in SOA [18]:
examine each of its existing systems, identify common
Bad selection of application infrastructure components,
processes/functions and identify services from these
(bottom-up). Either way identification of services is a Insufficient validation of the SOA enabling infrastructure
major issue. implementation,
Determine what legacy/existing components can satisfy SOA infrastructure, service and consumer application are
the service needs, determine what needs to be done with insufficiently instrumented for security, management and
those components and migrate and reuse them as services. troubleshooting.
In doing this there is a need to determine the cost and
effort of such a migration so as to make informed Many vendors IBM, Microsoft, SAP, Oracle, BEA, and many
decisions about the feasibility of mining and reusing. A others are advocating the use of their SOA infrastructure
method to help support determination of the cost and technology. Open issues in this area are how do organizations
effort is SMART [1] choose which SOA infrastructure to acquire and what pieces they
will need to build themselves? How des an organization test the
No matter which approach is chosen (bottom-up or top-down), if infrastructure sufficiently to guarantee the level of security,
legacy components are used to satisfy service needs it is important governance and management provided meets their requirements?
to identify services that meet performance and QoS requirements. How is interoperability between different vendor infrastructure
Approaches to identification and mining of services are needed that handled?
take these requirements into account and carry out a detailed
analysis of whether or not these services will satisfy current and
future QoS requirements.
2.4 Service Development and Application
Development from Services
As an organization moves forward with its SOA strategy it will need
2.2 Integration of Services to develop services as well as develop applications using either its
As an organization moves forward with its SOA strategy services home-grown services and/or services provided by external
will be identified, either internal or external, that will need to be organizations. Different approaches to developing services and
integrated with the organizations existing systems. Organizations applications can be used such as the traditional spiral development
will need to be able to determine what systems the services will models to more agile approaches. The approach used can also be
integrate with, the scope of the work involved and estimate the cost top-down which is driven by business processes or Model Driven
and effort of doing the work. The organization will also have to Architecture or bottom-up which is driven by existing legacy and
determine what the impact of different architecture alternatives will componentization of legacy [15].
be on various quality attributes and QoS concerns for the system. As
well as this it will have to determine SLAs for its applications and When dealing with the development of services and applications
system as well as what QoS requirements are need for the services from services one of the main concerns is dealing with the quality of
that it consumes and be able to negotiate these with the service service of the services and the quality of service of the application
providers. (which is dependent on the QoS of the services used). One of the
main issues is the problem of time-boxed development of services
We are currently developing a method to tackle these issues and are versus designing for QoS. If a set of services have to be developed
calling it SMAT-AUS. The method consists of several activities for in a particular timeframe especially in more agile approaches how is

36
QoS for the services handled (which may endure through the entire left to later? What areas can be automated and what tools are there
development process). to help automate governance? One of the areas of SOA governance
that we are working on is on Service Level Agreements and the use
When developing services and applications an organization will
of methods and tools to determine the performance and QoS of
have specific QoS requirements which the services and applications
services and applications that can be used in developing and
will have to meet. These requirements can guide the development of
negotiating SLAs.
the services and applications but there are few development
methods that can guarantee that the resultant services or applications
will meet their QoS needs. Detailed testing of the services and 3. PERFORMANCE AND QOS
applications will need to be done to determine if in fact they Since 2006 we have been researching, developing, and trialing
actually meets their QoS requirements. methods and tools for the performance modeling of SOAs. It is vital
to consider performance while architecting and developing SOA for
If external services are used in the development of an application both services and applications. Some example issues with SOA
then service level agreements with the service providers will have to performance are as follows. The performance (response time) of
be negotiated and be in place in order to make any guarantees about web services is typically worse than the bare underlying
the QoS level of the application. A further complication on QoS is functionality. This is due to and compounded by the overheads of
that an application may identify what services to use dynamically so XML (e.g. delivery, parsing, validation, serialization), the
it is not possible at design time to know what services will actually implementation of composite services (e.g. use of other services
be used. In this case it may be difficult to do QoS analysis of an including 3rd party services and legacy applications), service
application. In Section 3 we discuss our current work in the orchestration (infrastructure and protocols), service invocation (e.g.
performance and QoS area and outline areas where further research security, transports), resources (physical or software resources, e.g.
is needed. CPUs, threads), and resource models (e.g. virtualization). Optimally
In order to undertake a project to develop a service or an application allocating resources to services is also difficult, as resources may
from services there is a need to know the scope and size of the work not be correctly distributed, and individual services in a workflow
involved. This will help in determining the cost and effort for such a for which there is high demand, may become bottlenecks due to
project. There are few methods that assist in determining the scope, resource constraints. The consumers expectations of performance
size, cost and effort of SOA projects. Some guidelines as to what is may be higher than what can actually be delivered. For example, we
involved in determining the cost have been developed [16] but there have observed SLAs which promise a sub-second response time,
are few methods that do a complete scoping, sizing, cost, and effort whereas in practice response time ranged from 1 to 48 seconds.
estimation. More is needed in this area as having methods to Response time is often unpredictable, and varies depending on
determine the scope and reasonably accurate cost and effort figures service operation, binding, parameters, and payloads. If a call to a
will help organizations to better budget and plan for their major service times out unexpectedly the consumer is unsure of the
SOA efforts. success or failure of the call (e.g. if side-effects are possible).
Perhaps the most severe issue is scalability for Open services i.e.
2.5 SOA Governance services that are exposed publicly. The maximum number of
SOA governance is an essential activity to manage many levels of consumers, and the arrival rate and distribution are unknown in
SOA decisions correctly. There are various requirements to the advance, and extremely high peak demands are possible which may
governance problem including the following [17]: saturate one or more resources for the service resulting in increasing
response times, or more likely, the service being unavailable due to
Strategy and Goals what is governed and why? failure or the imposition of access restrictions to limit demand (load
Funding, Ownership and Approvals who owns what? shedding). For example, Open services may be used to construct
What gets funded and by whom? 3rd party mash-ups, resulting in unexpected increases and
distributions of demand for the consumed services.
Organization what structures, processes and governance
In the e-Government area we observe that there is increased
mechanisms are in place?
dependency between whole-of-government services, as
Processes what are the roles, responsibilities and compositions of services are offered and reused across all level of
procedures for managing SOA activities? government, non-government organizations, and business and
private citizens. The ability to architect applications with well
Policies what are the enforcement issues including known performance characteristics is critical, but becomes harder
standards, security, release and re-use? because of the dependencies between services, and the possibility of
Metrics what are the business outcomes and how are large numbers of users accessing applications during periods of peak
they measured and by what metrics? demand (e.g. to meet deadlines for submission of information to
government agencies).
Behavior what is the behavioral model, incentives,
penalties and rewards for appropriate SOA Behavior? Given that performance is a well known issue with SOAs, it is vital
to architect SOAs with performance in mind. To do this it is
Work still needs to be done in various areas of governance and clear necessary to be able to understand the performance characteristics
paths through all of the issues in governance have to be developed of the architecture and investigate alternatives early on in the
that will lead an organization to success with its SOA initiative. development lifecycle. To this end we are developing a method and
What governance issues are core to the success of an SOA initiative tool support for early lifecycle performance modeling of SOAs [3].
and need to be tackled early in the SOA lifecycle and which can be The tool supports performance modeling in terms of SOA concepts

37
(services, workflows, and composite services), and automatically 3.2 Virtualization
transforms the SOA model into a performance model which is Our SOA performance modeling method and tool support models
simulated producing performance metrics for the SOA components. services as first-order components. This makes the modeling easier
Significant performance modeling issues that we encountered and and more transparent for architects, and allows metrics to be
have made initial progress with are in the areas of service produced for services. Our tool was initially designed to model
orchestration (e.g. workflows and engines), resource virtualization resourcing of services on fixed resources. Simple service resourcing
(e.g. virtual servers), and Service Level Agreements (SLAs). We options are services deployed to non-shared servers (i.e. 1 service
believe that research in these areas is important but so far per server, a possible resource model for compositions using
incomplete, and we address each briefly in turn. services from different 3rd party providers), and services sharing
resources (i.e. multiple services per server, a typical resource model
3.1 Service Orchestration for compositions using services provided within the same
Composite services may be orchestrated in a variety of ways. For organization and hosted on common enterprise servers). However,
example, hardwired workflows over an Enterprise Service Bus, or one of the government SOAs we trialed the method with intended to
using a workflow language executed by a workflow engine. Each use resource virtualization in the form of virtual servers (e.g.
approach has different performance and scalability characteristics, VMWare, [7]), which requires modeling of services sharing
which depend on the workflows executed on them, their deployment resources dynamically (i.e. depending on maximum resources
and configuration, and the resources available. available, current resources available, contention for resources, and
resource sharing policies).
There are some academic studies of workflow performance [5, 6],
but little public empirical performance data is available from real Virtualization makes performance modeling more complex, but is
SOA workflow deployments. However, anecdotal information becoming an increasingly common foundation for SOA
suggests that workflow performance is a critical issue for SOAs. environments, so there will be a need to model increasingly
Workflow engines in Grid computing must discover and utilize complex virtualization scenarios. One of the significant issues is
large numbers of resources dynamically in order to support high gaining an adequate understanding of the performance impacts of
concurrency on-demand computations, and may need to be the VMware (for example) and various deployment, replication and
distributed for scalability. In Enterprise environments, the impact of configuration options available. For example, there is a fixed per
rules evaluated in a rule engine for the workflow steps may be VMware instance overhead, but scalability is harder to predict when
substantial. Configuring workflow engines for different levels of replicating VMware instances, or increasing the number of physical
reliability and failover can have a large performance impact (e.g. if CPUs per VMware instance (there are indications of non-linear
every step is persisted to a database, or if the state of replicated overheads [8, 9]). There is a lack of benchmarks showing the impact
workflow engines is synchronized to enable warm failover). on application performance of VMware configurations, as the
Understanding the performance characteristics and performance VMware-specific benchmark is designed to show benefits of
tuning of workflow technology is vital for the architecting and VMware for increased utilization of servers.
operation of production systems, and requires expert knowledge.
We have developed tool support for modeling one example of
Due to the architectural mismatches across services [4] and the
server virtualization. This is done by weighting the deployment of
resulting difficulty of integrating services that were not originally
services to physical servers, the weights representing virtual CPUs,
designed to work together, XML data must be transformed,
allowing the resource simulation model to allocate more resources
serialized, and/or validated between services (i.e. from the output of
to services with higher weights. However, more flexible modeling
one service to the input of another service). Transformation may be
of virtual servers is needed to model various scenarios of services
done in the workflow engine itself (e.g. by XSLT), or by specialized
deployed to software servers or containers (e.g. web servers,
integration infrastructure (e.g. Composite Application Frameworks).
application servers), software servers hosted on virtual servers with
It is often resource (computationally and memory) intensive.
virtual resource allocations (e.g. virtual CPUs), and virtual servers
Our SOA performance modeling method models service resourced by physical servers.
compositions as a series of workflow steps (a step is a service and a
The scope of virtualization is increasing in enterprises. Resources
time, either measured, or the default time of the invoked service, or
other than servers are now being virtualized, for example networks,
computed by the simulation model for composite services). We do
SANs, etc. This will make performance modeling of SOAs
not currently model the orchestration technology directly, as the
deployed to completely virtualized environments more challenging.
overhead of orchestration is included in the measured performance
times. In theory the orchestration overhead could be modeled The relevance of virtualization to SOAs has been documented
explicitly as time spent in a workflow service. However, these previously (e.g. for remote deployment of services across
times need to be measured for the target system, and a suitable organizational boundaries, [10]), and recently Amazon has
model of the performance characteristics of the workflow service introduced a virtual server based on-demand cluster computer
need to be developed for different orchestration solutions. Also the service (Amazon EC2, [11]), which is web service-based and uses
impact of other resource limitations may need to be modeled. For Xen, another virtualization technology. Another application of
example, memory usage due to large payloads held in memory for virtualization is to isolate consumers from each other to ensure
the duration for each workflow instance may severely limit the SLAs can be met. This is a logical extension of the Amazon
number of concurrent workflow instances. We are currently approach where each user has one or more exclusive virtual
exploring options to extend our tool to allow resources such as environments (and is only charged for the physical resources used).
memory to be allocated, held, and de-allocated across multiple See below for further discussion.
service boundaries.

38
3.3 Service Level Agreements and performance will start to degrade. Strategies to guarantee QoS
This brings us to the final issue, how can SOAs deliver what they include preventing SLAs being violated (e.g. by limiting the number
promise to different user communities (with possibly virtualized of concurrent connections per user community), or increasing
resources) [12]? As part of the trials we have assisted government resources dynamically (e.g. by using autonomic resource allocation
agencies understand the connection between the modeled in conjunction with virtualization).
performance of SOAs and SLAs. Because composite services can It is unclear how practical it is to deploy a service for each SLA to a
consume external 3rd party services, service providers function in virtual server, as it is not just the externally visible service that
the roles of both service consumers and providers, and therefore needs to be deployed to the virtual server, but also all dependent
need to understand SLAs in both contexts. If services are consumed, services. Finally, if SLAs are in place then they must be monitored
the performance model can be used to understand the demand for for enforcement, to locate the origin and cause of potential SLA
external services. If there is a SLA associated with the use of these violations (e.g. is the problem on the consumer, network, or
services, it is possible to determine if the demand will be exceeded, provider side) to allow for future capacity planning, renegotiation of
and the SLA renegotiated, costs or other penalties for exceeding the SLAs, and possibly for dynamic allocation of resources (e.g.
demands estimated, decrease in response time expected taken into resources can be easily increased in Amazon EC2).
account, or measures put in place to limit the demand (e.g. by
limiting the concurrency in a wrapper service for invoking 3rd party
services). 4. CONCLUSION
As organizations move to introduce SOA and develop SOA
From the perspective of service providers, it is possible to use the applications there are many aspects that they may encounter. In this
performance predictions to negotiate SLAs for consumers, or paper we have outlined some of those aspects and outlined some of
determine the resources required to meet previously agreed SLAs. the problems and issues that organization will face when tackling
Once the maximum throughput of each service is known, by using the various aspects. We have also outlined some open issues in these
the maximum predicted throughput as the average arrival rate for a areas where further research needs to be done.
Poisson distributed workload, it is possible to find the maximum
permissible average arrival rate to ensure that the system is not No matter which aspect of SOA migration that an organization is
saturated for some maximum period of time per day (or other unit of involved with one of the main areas of concern for SOAs is that of
time) with a given probability (e.g. less than 1 chance in 100 of performance and QoS. We have outlined some of the work that we
service having excessive response time due to resource saturation are doing in this area and in particular what are still some of the
for more than 10 seconds in 24 hours). For different classes of users main open issues related to performance and QoS. This work can
(e.g. different organizations) it is also possible to determine SLAs to especially be used in the SOA governance area when determining
split the available capacity between them by imposing limits on and negotiating SLA both for services (internal and external) and
their arrival rate for the service, along with expected response times applications developed using services.
(e.g. max average arrival rate of 1 every 5 seconds to ensure The research areas in performance and QoS areas that we are
response times below 3 seconds). Of course, this is based on working on; service orchestration, virtualization and service level
independent and uniform distribution of arrivals. If the arrivals are agreements are areas that are important to get right for an SOA
dependent or non-uniform (e.g. DoS attacks, burst of arrivals to system. The results of our research should lead to improved
meet deadline, etc), then all bets are off and the services may be modeling capabilities, better modeling and simulation tools and
saturated for significantly longer than estimated for all users (e.g. better use of the data produced by these tools.
record stock trading volumes caused a meltdown resulting in
services being unavailable for 30 minutes, [13]).
5. REFERENCES
It would be desirable to automate further the process of deriving [1] Lewis, G., Morris, E., OBrien, L., Smith, D., and Wrage, L.,
SLAs from the performance metrics, and allow various tradeoffs to SMART: The Service-Oriented Migration and Reuse
be made for different SLAs depending on the needs of the different Technique, G, CMU/SEI-2005-TN-029, Software Engineering
user communities. The other requirement is to determine the Institute, USA.
resources required to meet a set of SLAs.
[2] Zhang, Z. and Yang, H., Incubating Services in Legacy
A significant problem remaining however is the inability to isolate Systems for Architectural Migration, 11th Asia Pacific
the SLAs for each user community to ensure that overuse by one Conference on Software Engineering, Nov 20 Dec 3, 2004,
community doesnt degrade the agreed QoS for the others. One pp: 196-203.
approach that could be pursued is to use virtualization to isolate [3] Brebner, P., OBrien, L., Gray, J., Performance Modeling for
services for different user communities from each other. This would Service Oriented Architectures, research demonstration track,
allow a SLA to be specified for each virtualized service. While there ICSE 2008.
are spare resources, SLAs will be met, and even exceeded, as with
virtualization spare resources are distributed to active processes, so [4] Bierhoff, K., Grechanik, M., and Liongosari, E. S. 2007.
even if a user community is exceeding the agreed demand for a Architectural Mismatch in Service-Oriented Architectures. In
service they will receive the agreed QoS. Concurrently other user Proceedings of the international Workshop on Systems
communities will continue to receive the agreed share of resources Development in SOA Environments (May 20 - 26, 2007).
required to ensure that their SLAs are still met. International Conference on Software Engineering. IEEE
Computer Society, Washington, DC, 4. DOI=
However, once overall demand exceeds supply, resources are http://dx.doi.org/10.1109/SDSOA.2007.2.
strictly allocated according to the virtualization resource policies,

39
[5] Aiello, R., Workflow Performance Evaluation, March 2004, [14] OBrien, L., Moreno, P. and Bass L., Quality Attributes for
PhD, Universita di Salerno. Service-Oriented Architectures, International Workshop on
[6] Kim, K. and Ellis, C. A. 2001. Workflow performance and Systems Development in SOA Environments (SDSOA) 2007
scalability analysis using the layered queuing modeling at the International Conference on Software Engineering,
methodology. In Proceedings of the 2001 International ACM Minneapolis, USA, May 20th 26th, 2007
SIGGROUP Conference on Supporting Group Work (Boulder, [15] Arsanjani, A., Toward a pattern language for Service-Oriented
Colorado, USA, September 30 - October 03, 2001). (Ellis, C. Architecture and Integration, Part 1: Building a service eco-
and Zigurs, I., Eds. GROUP '01. ACM, New York, NY, 135- system, IBM, 2005, URL:
143. DOI= http://doi.acm.org/10.1145/500286.500308. http://www.ibm.com/developerworks/webservices/library/ws-
[7] VMware, www.vmware.com. soa-soi/

[8] VMware ESX Server 2: Performance and Scalability [16] Linthicum, D., How Much Will Your SOA Cost? Here are
Evaluation, 2004. URL: Some Guidelines, Info World November 2006.
ftp://ftp.software.ibm.com/software/dw/wes/hipods/esxserver_ [17] Marks, E. A., Perspectives on SOA Governance, URL:
evaluation.pdf. http://colab.cim3.net/file/work/SOACoP/2006_05_2324/EMar
[9] Improving Scalability for Citrix Presentation Server. URL: ks05242006.pdf
http://www.vmware.com/pdf/esx_citrix_scalability.pdf. [18] Gartner Inc., Bad Technical Implementations and Lack of
[10] Brebner, P. and Emmerich, W., Deployment of Infrastructure Governance Increase Risk of Failure in SOA Projects, Media
and Services in the Open Grid Services Architecture (OGSA), Release, June 2006. URL:
Third International Component Deployment Conference, CD http://www.gartner.com/it/page.jsp?id=508397
2005, Grenoble, France, Dearle, A, Eisenback, S. (Eds.), [19] Liu, Y., Zhu, L., and Gorton, I., Performance Assessment for e-
LNCS Volume 3798/2005, pp. 181-195. Government Services: An Experience Report, the 10th
[11] Amazon EC2 (Elastic Computing Cloud). URL: International ACM SIGSOFT Symposium on Component-
http://www.amazon.com/gp/browse.html?node=201590011. Based Software Engineering (CBSE), Lecture Notes in
Computer Science, Boston, USA, 2007.
[12] Ludwig, H., Web services QoS: external SLAs and internal
policies or: how do we deliver what we promise?, Fourth
International Conference on Web Information Systems Acknowledgements
Engineering Workshop, 2003, p. 115-120, Digital Object NICTA is funded by the Australian Government as represented by
Identifier: 10.1109/WISEW.2003.1286793 the Department of Broadband, Communications and the Digital
[13] SMH, Share panic leaves online traders in the lurch, January Economy and the Australian Research Council through the ICT
24, 2008. URL: Centre of Excellence program.
http://www.smh.com.au/articles/2008/01/24/1201025051004.h
tml.

40

Das könnte Ihnen auch gefallen