Community Presentations

October 3-6, 2005
Boston, MA, USA

Community Activity: Web Services Performance: Issues and Research
Tuesday, 04-October 11:00a – 3:30p

Kenneth Chiu and Yuichi Nakamura, SUNY Binghamton and IBM Tokyo Research, respectively

The format of this activity will consist of a series of 30-minute talks, plus a panel discussion.


Geoffrey Fox
Director, Community Grids Lab
Indiana University

Wolfgang Hoschek
Postdoctoral Fellow, Distributed Systems Department
Berkeley Laboratory

Michael J. Lewis
Associate Professor of Computer Science
State University of New York at Binghamton

Noah Mendelsohn
Distinguished Engineer

Rich Salz
Chief Security Architect, DataPower Technology

The convergence of Web services with Grid computing has brought significant benefits. Lingering performance concerns of Web services, however, threaten to slow the adoption of Grid services, and limit its penetration into the cyberinfrastructure. Research avenues that may contribute to addressing the performance issue include advanced parsing techniques, alternative binary encodings of XML, and streaming XML processors to address memory usage. This half-day workshop seeks to foster discussion of these issues by bringing together researchers and application developers, and also to discuss metaquestions such as whether or not performance concerns are valid, and if so, what is the impact.

The convergence of service-oriented computing and Grid computing has led to the wide acceptance of Web services as the standard communication mechanism in Grid computing. This has brought the benefits of improved interoperability in heterogeneous environments, and has allowed Grid computing researchers and developers to leverage the substantial efforts of the wider Web services community.

Performance, however, has been a lingering concern of Web services. Common implementations of Web services have developed reputations, deserved or not, for being slow and memory-intensive; or in some cases simply unable to handle large messages. Secure web services must also contend with XML canonicalization, XPath query computations, and cryptographic computations. The performance issues have been attributed variously to underlying causes such as the inherent verbosity of
XML, the costs of converting floating-point numbers from the native machine representation to a text format, and the proliferation of DOM-based trees in XML processing streams.

The perceived inefficiences have slowed the adoption of Web services, and limited its use to communications deemed not performance-sensitive. Performance sensitive data and communications are then conveyed using mechanisms, often ad hoc, proprietary, or binary-based, that are considered more efficient. This has two disadvantages. First, it prevents the benefits of Web services from penetrating deeply into the information infrastructure. For example, semantic mediation is a technique which can be used to address data compatibility issues, but works best with XML-based data. Data format description languages can be used to reconcile semantic mediation with arbitrary data formats, but introduces another layer of complexity.

Second, it prevents a single, standard data model and terminology from being adopted throughout. One set of concepts and models is required for Web services, and another is required for the high-performance technologies. For example, binary attachments can be used in SOAP to convey scientific data, but that scientific data then requires understanding another type system to interpret the contents. If Web services were efficient enough, sending the data directly in the Web service as XML may simplify the overall architecture and improve the interoperability of scientific data. These issues are increasingly important as we seek to improve scientific data management, provenance, and compatibility issues.

A number of different avenues of research may contribute to addressing Web services performance. Improved XML parsing may alleviate some bottlenecks. Streaming approaches may address other issues, such as memory footprint. Benchmark suites might help focus research and separate myths from reality. Alternate binary encodings of XML may address XML verbosity and numeric
conversion issues.

Goals of Session

The goal of this session is to foster discussion on these research directions, and metaquestions such as when Web services performance is a problem, the impact of Web services performance issues, and even whether or not Web services performance is an actual problem. We hope to draw an audience consisting of researchers in Web services performance, and application developers and architects who are developing production Grid systems.

Target Audience
The target audience for this workshop is architects, developers, and engineers of production Grid systems; and researchers in Web services performance. This includes the following communities:

- researchers and developers in XML/Web services performance, developers of parsing tools, XSLT/XPath performance optimisers, and SOAP engines

- researchers and application developers in Grid computing, utility computing and on-demand computing, especially those based on service-oriented software architecture and web service technologies

- engineers and developers from industries/businesses who think about using Web services or have used Web services but are concerned about performance or have experienced performance problems

- researchers and specialists in performance and measurement who have studied performance issues in Web services technologies

Potential Attendees
XML performance: there is a very active industrial and academic community of researchers and developers of XML performance tools and optimisers, including organiser Y. Nakamura. This group of people is a main target for participation, including people from IBM, Airbus, smaller software companies, and academia such as Karlsruhe, INRIA, University of Pennsylvania, etc.

Grid and other Web service application areas: the Global Grid Forum and industries such as HP and IBM heavily push service-oriented computing, but often have linger-ing doubts about performance of the solutions. We expect interest from researchers and developers of these cutting edge applications, including ondemand, utility, en-terprise integration and business-to-business applications, and organisor K. Chiu is closely involved with some specific application areas in which low overhead is critical. Since the range of applications is large, we target participants from various academic and industrial organizations.

Workshop Format
We are considering a 3-4 hour session consisting of about 6-8 invited presentations divided into two categories. Each category would be followed by a short panel discussion consisting of the presenters.

Non-Exhaustive Topic List

Any innovative and rigorous approaches, theoretical tools,
engineering methods, practical implementation analysis and
experimental reports related to the following topic list are of
interest for this workshop.

- XML parsing/validation
- XML transformation, e.g., XSLT, XPath, and XQuery
- Binary message representations
- Web services engines performance, e.g., message deserialization
- Security overhead, e.g. WS-Security and federation
- Enterprise service bus, mediation, brokering
- Grid computing performance, e.g., OGSI, WSRF
- Industrial experiences in Web services applications
- H/W acceleration
- Prediction, benchmarking tools and techniques


Kenneth Chiu, Assistant Professor of Computer Science, State University of New York at Binghamton

Kenneth Chiu is an assistant professor of Computer Science at the State University of New York (SUNY) at Binghamton. His research interests are in distributed systems, with a focus on web services and Grid computing. He is currently a co-PI on three
NSF-funded projects to develop service-oriented systems for scientific instruments and sensor networks. The first of these is part of the NSF Middleware Initiative, and the second of these is the first award of the NSF National Ecological Observatory
Network (NEON) project. The third project is the CrystalGrid project which seeks to integrate data management and instrumentation for X-ray crystallography. He received his undergraduate degree in Computer Science from Princeton University, and his Ph.D. in Computer Science from Indiana University.

Yuichi Nakamura, IBM Tokyo Research Laboratory, Japan.

Yuichi Nakamura is a manager of the Software Lifecycle team at IBM Tokyo Research Laboratory. His team is running Model-Driven Development, program analysis and end-user programming projects. His research interests are Web services technologies. Some projects he is conducting are found in One of his key accomplishments is the development of a Web Services Security (WSS) component for the IBM WebSphere Application Server product, as a technical lead. Currently, he is working on Web services performance issues, and usability of WSS tooling. He received his Ph.D. in Applied Physics from Osaka University in 1990.





Contact Webmaster.
OGFSM, Open GridForumSM, Grid ForumSM, and the OGF Logo are trademarks of OGF