Ultimately, VINT will enable more rapid design and evaluation of protocols and thereby directly contribute to the evolvability of the global information infrastructure.
This work is being performed in collaboration with LBNL, UC Berkeley, and Xerox PARC.
The composable component framework allows component modules from multiple contributors. This mirrors network development and is feasible because of the modularity inherent in the network protocols themselves. For example, the designers of a reliable multicast transport protocol could evaluate their design in the presence of simulation modules that represent dynamic multicast routing and packet scheduling algorithms. Or similarly, the designers of packet scheduling and queue management mechanisms would be able to evaluate their design in the presence of simulation modules that represent various subnet technologies (e.g., wireless).
In addition to creating the composable framework, the VINT project is also implementing and studying key protocol modules, including unicast routing, multicast routing, transport and web protocols. This will allow the evaluation and improvement of the simulator framework, and more importantly, will give other researchers a relatively complete ``internet context'' in which to plug their components for evaluation.
One of the most significant impediments to more extensive simulation use is the amount of effort typically involved in setting up and interpreting simulations. Graphical manipulation of inputs and animation of simulation processes and outputs should provide the same benefits to network simulation as they have to so many other areas of computer aided design and visualization.
In summary, the goal for the detailed simulator component of VINT is to enable extensive simulations by innovating in the area of simulator composability and visualization. However, such a system alone is not adequate because the detail necessarily limits the tractability of studying very large networks. Therefore, the second component of the system is included as described in the next section.
One of the greatest challenges, with respect to this component of VINT, is to develop methods for moving between detailed and high level abstraction: for example, creation of static scenarios at one level based on results at the other, and dynamic interfacing of detailed and abstract modules. In addition, tools for creation of session level protocol modules based on detailed simulation or implementation modules are being investigated.
Special visualization techniques will be needed for the session level simulator to allow graphical definition and manipulation of the very large networks, and visualization of their processes (intermediate states) and results.
In summary, session level simulations emphasize scaling issues in addition to multiprotocol composability. Development and application of validation techniques is being done by correlating session-level simulation results with both detailed simulations, smaller scale testbed experiments, and operational network measurement.
To achieve this last goal, early prototypes of VINT are being used in ongoing research efforts. This is providing initial experience with VINT and feedback to its design. These research efforts will explore interactions between component protocols. In particular the other areas of study are: the interaction of routing and end-to-end service; reliable multicast transport and underlying multicast routing; Web caching mechanisms and multicast routing; packet scheduling algorithms and end node application performance; impact of wireless MAC level protocols; congestion control and dynamic topologies.
Some of these issues require a detailed study of protocol interaction to evaluate their functional correctness; others require coarser grained evaluation to characterize performance issues. For example, a detailed protocol simulator and associated visualization tools are required to evaluate the functional interactions between the stack of protocols needed to support multiparty applications. Reliable multicast and error recovery mechanisms operate over multicast distribution trees established by the multicast routing protocols, which in turn depend on the unicast routing protocol; moreover, the protocols' behavior is affected by the presence of multiaccess subnets and must be studied in topologies with both these and point-to-point links.
A second set of examples involves the performance of web protocols. For instance, it has been proposed that the web-cache placement could make use of either network topology information provided by monitoring of network paths, by the routing protocol, or by a network management function such as the Internet Routing Registry. A session level simulator is being developed to evaluate the tradeoffs associated with the granularity at which such topology information is provided. On the other hand, a detailed simulator would be needed to study another frequently discussed performance issue associated with the World Wide Web, namely, the interaction between HTTP and the congestion control mechanisms being used by other types of traffic.
Initial results of protocol stress testing method by subjecting localized protocol mechanisms to complete error-scenario generation. Applied to multicast routing protocol.
Initial enhancement of network animation techniques (nam2) to capture link, node, and protocol state dynamics.
Evaluations of performance bottlenecks in simulation environment--memory and runtime.
Development of a prototype session level simulator with common interfaces, protocol modules, and a network protocol description interface. Initial target application is reliable multicast protocol design.
Phase 1 Visualization tools for both detailed and session level simulators.
Results of protocol studies completed using VINT/ns, including the behavior of reliable multicast protocols in the presence of dynamic topologies, robustness testing methods for soft-state protocols, web caching mechanisms and underlying multicast delivery, as examples.
Significant effort is being focussed on tool support for reliable multicast protocol design with the hope of engaging the IRTF reliable multicast group to use VINT/ns as their evaluation platform. This will allow the evaluation of one of the ultimate goals of the project, namely promoting VINT simulations as a community standard for comparisons in published work. Similar efforts will be made in the context of multicast routing and queue management and scheduling mechanisms.
VINT/ns is already a common evaluation platform for TCP research, and we are continuing additions to VINT/ns to support that community.
The software availabe from the project are:
Tcl-v7.6, Tk-v4.2, OTcl-v0.96 and the latest version of libTcl available from the ns home page need to be installed to run the software.