Presentation on theme: "Multipoint Communications in a Beyond-3G Internetwork Elias C. Efstathiou & George C. Polyzos Mobile Multimedia Laboratory Department of Informatics Athens."— Presentation transcript:
Multipoint Communications in a Beyond-3G Internetwork Elias C. Efstathiou & George C. Polyzos Mobile Multimedia Laboratory Department of Informatics Athens University of Economics and Business Athens 10434, Greece Tel.: , Fax:
Outline l Introduction u The Internet Beyond 3G u Mobile Multicast: High-level Issues l IP Multicast, Mobile IP and Cellular IP l Filters and Media Stream Quality l Combining IP Multicast and Mobility u IETF Mobile Multicast Approach u Extensions to the IETF Approach l Our Perspective u IGMP Mobility Support and IGMP Assumptions u Multicast Semantics and Mobility u Mobile Multicast Requirements u Cellular IP and Mobile Multicast u The Beyond-3G Environment l Conclusions
Introduction - The Internet Beyond 3G l Diverse network technologies u 2.5G and 3G networks n 700 million cellular subscribers today n 2 billion expected in the timeframe u Digital Video Broadcasting (DVB) networks n Terrestrial (DVB-T), Satellite (DVB-S) flavors n 30 Mbps of shared downlink bandwidth u IEEE networks n 11 up to 54 Mbps in the ISM band n New business models emerging u Traditional wired access networks n Ethernet n PSTN, ISDN, DSL, Cable
The Internet Beyond 3G (cont’d) l Goal: All-IP internetwork u IP-over-everything, IP-under-everything u Integrated services spanning network technologies n Audio, video, data n Unidirectional and bidirectional u Support for Multipoint Communications? n One-to-many, many-to-many u Support for Mobility? n Personal mobility n Network mobility (moving ships, trains, cars) l Our focus: The Mobile Multicast Problem u IP-based quasi-reliable mobile multipoint communications n assuming a fixed routing infrastructure (no ad hoc networks) n assuming IPv4, but taking IPv6 into account
Mobile Multicast: High-level Issues l IP Multicast u Easy for some technologies only n “Native” support in Ethernet and broadcast networks… n …but, point-to-point links in Cellular, PSTN/ISDN and DSL l Mobility u Not supported in the original Internet design n An IP address is a subnet and interface identifier but it’s also used in packet routing n TCP connection identifiers include lower layer IP identifiers and do not allow them to change u Answer: Mobile IP n One address for identification, another for routing l Hostile Wireless Environment n Higher BER challenges original TCP and IP assumptions
IP Multicast l Many-to-many communication protocol u Host group service model n a receiving host may join and leave a multicast group at any time n all IP hosts can communicate unidirectionally with all group members using only the group’s identifier (its class D multicast address) u Routing packets and tracking membership n Global and Local mechanisms respectively l Global Routing Mechanisms u Multicast routing protocols deliver a group’s packets to multicast routers that have expressed interest in receiving packets for a particular group n DVMRP, CBT, MOSPF, PIM u Graft delay when a multicast router joins the multicast tree
IP Multicast (cont’d) l Local Membership Tracking Mechanisms u Multicast router: the “interface” between the local and the global mechanism n Exposes aggregate list of groups all its hosts have joined u Protocols for membership tracking n IGMP (for IPv4) and MLD (for IPv6) u Soft-state principle – no explicit LEAVE_GROUP primitive u IGMP assumes link-level native broadcast
Mobile IP (M-IP) l Allows internetwork host mobility in a manner transparent to the transport layer l M-IP terminology n Mobile Host (MH) n Correspondent Host (CH) n Home Agent (HA) n Foreign Agent (FA) n Care-of Address (CoA – 2 types: FA CoA and co-located CoA) n Tunneling (IP-in-IP encapsulation) l Registration with the HA needed n Soft-state principle, re-registrations required n Delays, lost packets l “Triangle routing” inefficiency n Solved in M-IPv6, which specifies that all CHs can make a CoA-to- home address binding and can also tunnel and de-tunnel packets
Cellular IP (C-IP) l C-IP is a micro-mobility protocol u Unlike M-IP’s “slow” macro-mobility, C-IP assumes “fast” mobility u Needed because M-IP incurs delays n FA discovery time + Registration with FA + Registration with HA u M-IP can violate “the mobility assumption” n “total registration delay maybe more than the time a MH spends inside a cell [controlled by one FA]”. No packets will be delivered l C-IP assumptions u Campus-wide, partially-overlapping micro-cells u Two-tier architecture: C-IP will rely on M-IP n Simpler than proposed “hierarchical FA M-IP schemes” (Realistic!) u C-IP gateway M-IP FA n “interface” between the M-IP and C-IP routing infrastructures u Base stations IP packet forwarders n Simple routing: no tunneling, resembles MAC bridge frame forwarding with auto-learning
Cellular IP Access Network Mobile IP Internetwork Correspondent Host Home Agent Foreign Agent and Cellular IP Gateway BS1 BS4 BS3 Mobile Host BS2 Cellular IP Network
Filters and Media Stream Quality l Filters and transcoders u Absolutely necessary for mobile multicast u Help maintain a level of Perceived Quality of Service (P-QoS) u “Smart” filters, “Simple” filters l Layered coding and multi-resolution layered coding u A media stream is separated into more than one stream n Sub-streams can be transmitted in different multicast groups n Receivers “tune into” as many as possible l Filter mobility characteristics u Fixed n Usually located at the boundary between wired and wireless section u Mobile n In multicast trees, they can propagate upstream, closer to the source, combine into one and serve many receivers in the same sub-tree
Combining IP Multicast and Mobility l Mobile devices are fundamentally different u Limited battery life have to avoid unnecessary operations n constant network traffic monitoring is impractical u Radio interface cannot assume high bandwidth nor low BER u Handoffs forced disconnections n Vertical and horizontal u TDMA and CDMA with power control no link-level multicast yet l IPv4 address shortage u GPRS operators rely on NAT n NAT makes IP multicast more difficult l Cellular operators interpret “multicast” differently u Cell-limited usually u Not the IP-based multicast envisaged for the Beyond-3G Internet
IETF Mobile Multicast Approach l RFC 3220 (Mobile IPv4) proposes two methods l (1) MIP-RS – “Remote subscription” u Assumption: a multicast router exists in the visited subnet n MHs simply use IGMP and (re-)subscribe to any number of groups u Disadvantage 1: delay, packet losses, tree rearrangement u Disadvantage 2: “get-ahead” and “lag-behind” problems l (2) MIP-BT – “Bi-directional tunneled multicast” u Assumption 1: the MH’s HA is a multicast router n IGMP requests are tunneled to the HA n The HA joins groups on MH’s behalf u Assumption 2: MH must decapsulate the multicast packets sent to it through the tunnel - even if it uses an FA for M-IP decapsulation u Disadvantage 1: potential packet duplication u Disadvantage 2: potential tunnel convergence
Packet Duplication and Tunnel Convergence FA MH HA FA MH HA Both visiting MHs belong to the same home network and are members of the same multicast group but because of tunneling the FA has no way of knowing this Visiting MHs belong to different home networks but they are members of the same multicast group causing multiple HA-FA tunnels to carry the same datagrams
Extensions to the IETF Approach l Mobile Multicast (MoM) Protocol n Based on MIP-BT. Key extension: The Designated Multicast Service Provider (DMSP) n A DMSP for a group is an HA chosen by a subnet’s FA out of the many that may forward packets for a specific group there n FA chooses a DMSP and performs DMSP handoffs when needed n Solves tunnel convergence n Most cited alternative to MIP-BT and MIP-RS n V. Chikarmane et al., “Multicast Support for Mobile Hosts Using Mobile IP: Design Issues and Proposed Architecture,” ACM/Baltzer Mobile Networks and Applications, 3(4): , Jan l Mobile Multicast with Routing Optimization (MMROP) n Based on MIP-RS. Key extension: The Mobility Agent (MA) n MAs route missing packets (via tunneling) to neighboring subnets n Tunnels need to be setup between FAs n Solves get-ahead problem n MMROP assumes packets are somehow numbered n J. Lai et al., “Mobile Multicast with Routing Optimization for Recipient Mobility,” Proceedings IEEE ICC2001, pp , June 2001.
Extensions to the IETF Approach (cont’d) l Constraint Tree Migration Scheme (CTMS) n Improved version of the CBT multicast routing protocol n “automatically [migrates multicast trees] to better ones while maintaining the QoS guarantees specified my mobile users” n Reduces packets losses due to reconfigurations and join delays n Difficult to deploy m Most multicast routers still run DVMRP n K. Chen et al., “CTMS: A novel constrained tree migration scheme for multicast services in generic wireless systems,” IEEE JSAC, 19: , October l Multicast Scheme for Wireless Networks (MobiCast) n Based on MIP-RS. Key-extension: The Domain Foreign Agent (DFA) n DFAs serve many small adjacent cells n Small cells are organized in one Dynamic Virtual Macrocell (DVM) n Similar to hierarchical FA M-IP and to our joint M-IP/C-IP solution n C. Tan and S. Pink, “MobiCast: A Multicast Scheme for Wireless Networks,” ACM MONET, 5(4): , 2000.
IGMP Mobility Support & IGMP Assumptions l IGMP was designed with Ethernet in mind l IGMP is not suitable for routers with point-to-point links u IGMP queries have to be issued to each one of these links u Not everyone will hear responses… u … unless the router multi-unicasts them u More state information needed at the router l IGMP is not suitable for mobile hosts u Mobile hosts cannot constantly monitor network traffic u Mobile hosts should not be forced to resend unnecessary data u Solution use explicit JOIN_GROUP and LEAVE_GROUP primitives
Multicast Semantics and Mobility l Multicast semantics require reexamination in the presence of host mobility l Example: consider two Ethernet IP subnets X and Y u Some MHs from X are visiting Y and some MHs from Y are visiting X u is the special IPv4 link-local all-hosts multicast group u A packet addressed to is sent to X – what happens? n Packet is delivered only to hosts in subnet X regardless if they are visitors or not n Packet is delivered only to hosts in subnet X that are not visitors n Packet is delivered to all X hosts irrespective of location u Answer: it depends on the originating service protocol l IPv6 will help: defines link local, site local and organizational local multicast scopes
Mobile Multicast Requirements l Significant vs non-significant moves n If a MH move causes the new subnet’s multicast router to subscribe to new groups, the move is significant n Non-significant moves should have no effect on the global mechanisms n Both types must appear similar from the user’s perspective l Multicast packet buffering n Buffer packets until when? n Disconnections due to m Handoffs m Physical layer problems m User intent l Mobile subnets n Deal with them as one logical entity l Roaming n Sophisticated authentication and pricing schemes are also required
Cellular IP and Mobile Multicast l Integrate efficient multicast mechanisms into C-IP u Use C-IP in conjunction with M-IP u Based on IETF’s C-IP and M-IP interoperability ideas u Scalability concerns l Similar to the MobiCast scheme u MobiCast DFA C-IP gateway u MobiCast DVM C-IP subnet l Based on MIP-RS u Closer to real-life network deployments n Campus-wide internetworks n UMTS cells n DVB-T macrocells u MIP-BT tunneling not scalable u MHs should first exploit resources in their immediate environment
Cellular IP and Mobile Multicast (cont’d) l “Flat” address space in C-IP n Multicast addresses do not appear different from unicast addresses n Simple mapping of IP identifiers to forwarder ports l C-IP has keep-alive mechanisms similar to IGMP n Adapt the C-IP “route-update” packet mechanism n MHs send these packets but instead of using their own IP address in the source field, they use the multicast group address instead n When, with C-IP forwarding, they reach the C-IP gateway, the gateway (M-IP FA) may then subscribe to a group, if it’s not receiving it already. n No need for IGMP. Reuse C-IP soft state mechanisms. l Multicast groups can be “virtual” C-IP hosts n C-IP forwarders should handle IP address-to-multiple ports mappings n C-IP gateway must be a multicast router n MHs use their IP stack in an “unconventional” way n Packet duplication only when paths towards receivers diverge!
Cellular IP and Mobile Multicast (cont’d) Global Multicast Routing Protocol transmission source Foreign Agent and Cellular IP Gateway BS1 BS4 BS3 Mobile Host BS2 Mobile Host The 3 MHs subscribe to by sending route_update with as the source IP address. BSs update tables accordingly for virtual MH “ ”: BS1 BS2 BS2 BS3 and BS4 BS3 and BS4 do link-local broadcast The FA/CIP-GW grafts to new multicast trees every time a route_update with a new group arrives. IGMP-like soft-state assured through C-IP mechanisms.
The Beyond-3G Environment l Many, superimposed, cellular technologies n Even if we ignore satellites… n DVB-T (1-100 Km) macrocells m DVB-T with 3G as the return channel m 5-30 Mbps of shared bandwidth n 3G neighborhood size cells n microcells l Devices with multiple interfaces n combined with DVB or GPRS l Mobile IP an accepted standard n M-IP support in devices and networks l Improved TCP versions l Filter and transcoding standards
Conclusions l The Multipoint Communications Problem u Standard IETF protocols can solve it n IP multicast n Mobile IP n Cellular IP u The Environment Beyond-3G will be based on IP u Therefore, most of these IP-based solutions will be applicable l Future Research Areas in Mobile Multicast u Ad hoc networks u Strong reliability u Security u Roaming and pricing agreements