Ted Hanss Applications Lead Internet 2 Project (on loan from the University of Michigan) 519 W. William St. Ann Arbor, MI 48103 ted@umich.edu Phone: 313.763.6185 Fax: 313.763.4434 Internet 2 Application Issues The Internet 2 Project mission is to "facilitate and coordinate the development, deployment, operation, and technology transfer of advanced, networked based applications and network services to further U.S. leadership in research and higher education and accelerate the availability of new services and applications on the internet". This consortium effort is investing in upgrading campus and national network platforms for such application areas as digital libraries, collaboration environments, tele-medicine, and distance-independent instruction. The emphasis is not on data rates, but on quality of service guarantees enhancing current applications and creating applications that were not previously possible. Internet 2 (I2) will ensure, on an end-to-end basis, that applications have the network resources they require. However, the broadest goal goes beyond building a network or deploying applications---we seek to establish a distributed knowledge system for achieving innovations in research, teaching, and learning. Tele-immersion, for example, could accelerate the discovery of a cure by a distributed community of medical researchers. This white paper surveys issues that the I2 application effort intends to address. The intention is that further discussions establish priorities and result in partnerships across education, government, and industry addressing areas of joint concern---turning I2 issues identified below into shared NGI efforts. An I2 deliverable is a set of year-by-year assumptions for when certain features will be available for applications exploitation. Establishing that list requires a dialogue among a wide community. Quality of Service A necessary investigation is defining user requirements for quality of service. Audio/video services require bandwidth reservation and loss rate and jitter guarantees, among other attributes. Electronic commerce adds transactional guarantees. Collaborative applications demand real-time services, such as virtual reality environments with sensory feedback. Digital libraries may exploit low-latency by creating new services, such as interactive relevance feedback. Dependability and reliability are critical quality parameters for all applications. Is "best effort" versus "guaranteed QoS" a binary choice or a graduated scale? The latter provides greater flexibility, but perhaps differentiated service levels need to evolve over time given current router performance. Network-Aware Applications Today, applications are typically not aware of any attributes of the underlying network. Can we create network-aware, adaptive applications that adjust their functionality as network conditions change? How do applications determine they are getting the QoS requested? As we assume IP as a common bearer service, what planned IP extensions can applications exploit (e.g., RSVP, IPv6) and what further extensions must be made? Stable APIs must be defined to assist application developers in accessing necessary QoS features. High Fidelity Audio and Video Standards-based transport services for streaming audio and video are emerging. However, a proliferation of proprietary encoding schemes will lead to needing different client packages to access content from multiple sources. In addition, proprietary codecs are typically targeted at dial-up connections. We can explore non-proprietary approaches to encoding video/audio content at high data rates. Digital libraries can then increasingly integrate audio and video with their text and image materials, which will serve as a basis for research into organizing and accessing multimedia collections. With high fidelity, multi-channel audio, distance-independent music instruction can carry the nuances of a jazz artist. Integrate other QoS attributes and the music instruction environment encompasses collaboration tools for shared composition and perhaps even distributed ensemble playing. Tools Many of the utilities we use today were developed with assumptions of congested networks with small MTUs. Researchers using very large data sets have identified improvements in FTP, for example, as a place where a relatively small investment could result in large productivity gains. Collaboration Infrastructure To build effective collaboration environments, we must address mediating control among participants, displaying arbitrary content on any screen with whiteboard overlay, synchronizing data streams, and session record and playback. Industry may address some of these challenges. However, the higher education community could focus on its interests, including integrating control of remote scientific instruments. Infrastructure Components I2 is not intended to solve current internet problems. However, barriers remain to deploying advanced applications. Security is a critical distributed systems component, along with directory services and electronic commerce. Inter-realm authentication is needed to deploy digital library applications accessible across institutions. In addition to the technical challenges, there are policy issues for which I2 could be a testbed---What are the most appropriate costing and charging schemes for reserving particular QoS attributes? Exploiting GigaPoPs The topology of the I2 network includes GigaPoPs, which serve as regional aggregation points. GigaPoPs improve traffic efficiency, for example, through local packet exchange. But they also provide a location for caching or replication servers for web, database, and file systems access. They may also provide inter-modal connections; the envisioned connectivity mesh will eventually include wireless links. GigaPoPs could host local down- and up-link satellite facilities providing economical, nation-wide multicast delivery of telemetry data, audio and video streams, database and software updates, etc. Scaling A distributed environment with international applications connectivity must support hundreds to thousands of simultaneous users per application and millions of users overall. Thus, we must undertake application-level modeling efforts with such inputs as user behavior, quality of service attributes, and caching strategies. The modeling will consider the number of simultaneous users per application and the number of simultaneous applications per campus and per GigaPoP. User Interfaces The above issues focus on exploiting lower layers for enhanced applications functionality. The user interface, though, will be the critical test of success. Do users establish their own preferences for QoS? If so, how is it presented? What are the user perception issues for dynamically adjusted QoS? Information visualization research is key to deploying enhanced digital libraries. Collaboration-enabled applications require investigations into shared work tools, while tele-immersion adds the challenge of how to simulate realistic shared presence. Conclusion By exploring these issues in a production-focused environment, we hope to identify the best practices that meet the needs of the higher education community and that lead to effective technology transfer. That diffusion process will allow us to move on to the next challenge, Internet 3.