Collaborative Science Portals

Investigator: G. Fox (with collaboration with NCSA Alliance team led by Dennis Gannon)

While at Syracuse, Fox has built substantial experience in building both collaboration systems and web-based problem solving environments (PSE). The base technologies are built around distributed objects with the collaboration system Tango Interactive using custom shared Java and JavaScript shared objects and the PSE’s being built with both CORBA and custom Java servers. This work will continue after his move to Florida State but this move allows one to opportunity to rethink the architectures and implementation from scratch. We will also integrate the existing CSIT expertise in PSEs (van Engelen) and Visualization (Banks, Erlebacher) into our new approach. For the latter, we will build on some initial work at Syracuse on collaborative visualization and use these ideas to integrate the CSIT Power Wall collaboratively into portals.

Problem Solving Environments (PSE) have been pursued for many years with the work at Purdue pioneering many important concepts. The increasing power of computers and the increasing capability of distributed object and web technologies are making this approach increasingly attractive for users and system builders. One uses the “Object Web” (CORBA, COM, Java, XML etc.) and a browser based user interface to provide a single integrated view or portal to the resources and tools needed by the scientist. In his research at CSIT, Fox will start with the successful Gateway and WebFlow systems developed at Syracuse and applied to several DoD and NSF projects. These have a classic three-tier architecture with client, brokers/servers and services in the three layers. High performance is obtained even while using Java and CORBA in the middle tier, by careful separation of control and data. The middle tier provides a flexible control layer implemented with proxies and traditional high performance mechanisms such as Globus and MPI are used for data transfer in the backend. This WebFlow distributed object technology has a powerful dataflow and coarse grain object computing model with all interfaces defined in XML and compatible with community activities. In this sense it is more powerful than the earlier NILE system while it is less ambitious than the Common Component Architecture, Legion and POOMA, which provide a fine grain, object model which we are exploring in the ComponentMesh project. Gateway has many features that will be preserved in our new PSE’s such as consistency with commodity standards (CORBA and XML). However we intend to work in the approach of the NCSA Alliance where a PSE is viewed as a Portal which has a set of services that combine those of well known web sites (such as Yahoo) with classic Scientific PSE’s. Further user customization allows one to support different views of the information and services and support both computational science and education uses. Systematic use of XML with well defined interfaces for both client/broker and broker/backend service seems key. Further relevant community standards are being set by the Grid Form (http://www.gridforum.org) and the computing portals spin-off from the Java Grande Forum (http://www.computingportals.org). We will implement early prototypes of the new portal approach on Gateway but expect that one can better use one of the emerging set of “Object Web operating systems” such as E-Speak or Ninja as the basic framework.

We have found a mix of success and failure with initial collaborative systems such as Microsoft NetMeeting, NCSA’s Habanero and Syracuse’s TangoInteractive. Higher speed networking and quality of service will address some of the difficulties such as variable quality in digital audio video conferencing; here we track the ANL/NCSA Alliance Access Grid project. We have been quite successful in educational applications but have yet to develop collaborative computing applications, which are both robust, and of compelling value. We will use the existing collaboration systems in early experiments but we intend to build much of the collaborative infrastructure from scratch replacing custom protocols and services by more robust systems developed for large-scale application areas like e-commerce. The collaboration service in science portals must support both asynchronous interactions, and real-time synchronous collaboration and mixtures thereof. Our web-based PSE approach implies that collaboration is a service that shares web-based distributed objects. However we also need to support several collaborative modes; shared display and both collaboration-aware and collaboration-unaware shared event models. Previous systems have focused on one of these mechanisms and have not been able to support the needed range of collaboration. Initially we will support these different modes with separate subsystems but will replace this by an integrated system CPW (Collaborative Portal on the Web) based on a generalized shared queued event service. This terminology indicates that our approach is to build first a portal with collaboration as one its services; this will be implemented using XML systematically to define the details of the collaboration and the portal infrastructure (e.g. Ninja’s event service) as the building blocks of the collaborative system. We believe this will integrate collaboration directly into the scientific analysis and make it more useful than before.

The multi-tier structure of Science Portals fits well with the resources to be added to CSIT. We need multiple back end compute servers (Teraflop machine, cluster and experimental parallel machines) represented by the brokered by Java/CORBA servers which are naturally implemented on the Information Infrastructure. Collaborative visualization and mobile user support is an important aspect of portals and this part of the resource will enable such research.