Thoughts from networking Many-to-many multicast In general the simulation network could be capability-no overlay network requirements of M&S ly reliable/real-fifme multicas Need a capability for end-to-end network status s it will be support M&S this also can be pone in an overlay network networked group communicatio non-multicast andardize on over network layer Missing/problematic critical middleware Implementation Questions NTP and/or GPs will real-tim ed to network time for XMSF s authentication/a and can be used to weateieve that Grid and Cluster style network computing will accommodate XMSF without modifications A ry ust be adequately funded for operation What We Believe is Available What We Believe is(Achievable s and multicast provided on P private-network Dos on a multi-network basis (not Internet W the problem is the business case, not the technology Multicast through overlay networks latency under 100 ms one-way in North amenica . VPN ing application-tra nt multicast Enhanced perfo for digital ibi through caching by Good global synchronization via NTP/GPS apply to dynamic d changed by simulations Reliable multicast for bulk data transfer9 XMSF Strategic Opportunities Symposium 6 September 2002 49 Many-to-many multicast n trend is away from providing this as a network layer capability - no good business model w one-to-many may become available n must define requirements for reliability * w e.g. selectively reliable/real-time, fully reliable/non-real-time w it is impossible to have fully reliable/real-time multicast n identifying and responding to congestion is a requirement n it will be necessary to support M&S needs for networked group communication over non-multicast network layer * XMSF Strategic Opportunities Symposium 6 September 2002 50 Thoughts from networking community In general the simulation network could be an overlay network * n for example, virtual private network (VPN) n allows an ISP or the Internet to meet specialized requirements of M&S Need a capability for end-to-end network status & performance monitoring * n this also can be done in an overlay network Standardize on over-the-net protocols n riding over standard Internet protocols n proven basis for enabling interoperability XMSF Strategic Opportunities Symposium 6 September 2002 51 Missing/problematic critical middleware n real-time object request broker n authentication/authorization services n real-time directory services n group coordination/synchronization n session coordination likely is not a problem w Session Initiation Protocol does signaling w automated setup/teardown still needs attention XMSF Strategic Opportunities Symposium 6 September 2002 52 Implementation Questions NTP and/or GPS will be needed to provide synchronized network time for XMSF n GPS is more accurate and can be used to synchronize a local NTP master We believe that Grid and Cluster style network computing will accommodate XMSF without modifications n as long as network capacity is sufficient A dedicated and monitorable test environment would accelerate development of an XMSF community n use Next Generation Internet networks (Abilene, DREN, etc.) n must be adequately funded for operation XMSF Strategic Opportunities Symposium 6 September 2002 53 What We Believe is Available QoS and multicast can be provided on a private-network basis (probably in NGI) Performance available off-the-shelf: n individual flows to ~100 Mbps n latency under 100 ms one-way in North America n jitter manageable by buffering, increases latency ~10% n packet loss <1% High performance end-to-end with instant startup is practical as long as reliable delivery is not needed n TCP does not scale well to wide-area flows above 100 Mbps Good global synchronization via NTP/GPS n secure NTP may be required in some cases XMSF Strategic Opportunities Symposium 6 September 2002 54 What We Believe is Achievable QoS on a multi-network basis (not Internet wide) n the problem is the business case, not the technology Multicast through overlay networks n VPN n middleware providing application-transparent multicast Enhanced performance for digital libraries through caching n individual flows ~1 Gbps by localizing access n does not apply to dynamic data exchanged by simulations Reliable multicast for bulk data transfer