VINT: Status and Plans Deborah Estrin: Project overview Steve McCanne: ns architecture John Heidemann: scaling, visualization Audience: Comments and questions

Contributors USC/ISI: Deborah Estrin, Mark Handley, Padma Haldar, John Heidemann, Ahmed Helmy, Polly Huang, Satish Kumar, Kannan Varadhan, Haobo Yu, Ya Xu, Daniel Zappala LBNL: Kevin Fall, Sally Floyd, Van Jacobson UCBerkeley: Elan Amir, Steve McCanne, Teck- Lee Tung, Daedelus group Xerox PARC: Sandeep Bajaj, Lee Breslau, Scott Shenker other ns users Funded under BAA

VINT Goals Facilitate protocol design, evaluation, comparison –Protocol studies and interaction –Scale (currently many thousands, goal of 50K) –Visualization –Methods –Emulation Encourage design progress through –comparable simulations –code/protocol module sharing –visualization support –scaling mechanisms that allow validation across abstraction levels –common techniques

Existing infrastructure Protocols: –TCP, SRM – unicast routing, multicast routing, –MAC, wireless –Router mechanisms –Admission control Dynamic topologies, loss models, source models Sessionsim for scaling nam network animation for visualization Emulation agents Significant use: - Sigcomm papers - rm, end-to-end research groups - ns-users from ~130 sites

Observations Visualization is powerful (essential?) for debugging Importing implementation code segments wholesale is problematic when investigating protocol designs Scripting language and fine grain composition powerful but also add complexity Being our own and each others users continues to be essential Validation tests essential for collaborative/contribut ed code base

Scaling What we have –Centralized unicast and multicast routing with and without dynamics –Session sim packet distribution with TTL and loss dependencies –Very large simulations for regular topologies What we will have –Cloud abstraction –Hybrid simulations –50K node support –Explore distributed simulation

Visualization What we have –nam v2 –node state as well as pkt animation –examples of use for visualizing measurements as well as protocols –Integrated graph presentation –LAN visualization What we should have –Sessionsim visualization –Summary techniques for large detailed simulation –Programmability by ns users –Interaction between ns and nam –Possibly exploit power and flexibility of Lucent tool kit?

Systematic Evaluation Methods What we have –Demonstration of STRESS –Validation tests –Some ad hoc approaches to topology generation and member placement What we should have –Scenario generation tools/templates for TCP Reliable mcast Mcast congestion Multipoint apps/protocols Mcast routing Unicast routing –Topology, Membership, and Traffic