Title: Explicit Multicast (Xcast) MD06 Yuji Imai Fujitsu SGM Rick Boivie IBM CLM Dirk Ooms Alcatel
1Explicit Multicast (Xcast)MD06 Yuji Imai
FujitsuSGM Rick Boivie IBMCLM Dirk Ooms
Alcatel
2Multicast applications
- Broadcast-like (one-to-many)
- Multicast of IETF meetings
- Distribution of broadcast TV channels
- Narrowcast-like (few-to-few)
- IP Telephony with conferencing
- Video conferencing
- Real-time collaborative applications
- Multiparty networked games
- Other applications with millions of concurrent
groups
- Todays multicast schemes support the first case,
but there may be a better solution for the second
case - 'It is believed that a "one size fits all"
protocol will be unable to meet the requirements
of all applications' - - as the Reliable Multicast Transport Group
said on their web page
3Cost of Multicast (1)
- Cost of Host Group Model
- multicast addresses must be globally unique gt
multicast address allocation servers required
MAAS (multicast address allocation servers),
MADCAP (client-server), AAP (intradomain), MASC
(interdomain) - Cost of Multicast Routing Protocols
- connection state state per group or source-group
in every node of the tree - signaling to create this connection state
- source discovery mechanism
- network wide announcement of core node (PIM-SM,
CBT) - floodprune nature (DVMRP, PIM-DM)
4Cost of Multicast (2)
Cost of current multicast per member
OK
costly
Number of members
alternative xcast
5Cost of Multicast (3)
- New multicast protocols (Simple Multicast,
Express, PIM Source Only) avoid some of the costs
(e.g. address allocation), but they still need
state and signaling per flow. - Our point is that just as
- Core routers dont keep track of individual TCP
flows today - Diffserv suggests that core routers shouldnt
have to keep track of large numbers of individual
QOS flows tomorrow - Its our belief that core routers shouldnt have
to keep track of large numbers of individual
multicast flows either.
6Explicit Multicast (xcast) basics
- "Multiple Destination Option on IPv6",
draft-imai-mdo6-01.txt - "Small Group Multicast", draft-boivie-sgm-00.txt
- "Connectionless Multicast", draft-ooms-cl-multicas
t-01.txt - "Distributed Core Multicast", draft-blazevic-dcm-m
obility-00.txt
- explicit list of destinations
- forwarding uses unicast route table
- complementary to current multicast model
- current multicast scales with number of receivers
- xcast scales with the number of sessions
- xcast can lessen the load on current multicast
7Xcast Advantages
- No multicast address allocation
- No multicast routing protocol needed (intra- nor
interdomain) - No state in routers
- No need for a core node (RP)
- Optimal route is taken (even when paths are not
symmetrical) - triangular routing avoided
- maximizes network efficiency
- Automatic and immediate reaction to unicast
reroutes - Simple (only data plane changes)
- Lends itself to Reliable Transport
- Destination aware simple security and accounting
- Simple transition schemes
8Xcast Basic Mode Disadvantages
- Overhead in packet header
- More header processing
- n routing look-ups
- on the fly header changes
- Doesnt support large groups
- gtcomplementary to current multicast
9Xcast Flexibility
gt xcast can adapt to specific network
constraints (access, backbone, LAN, )
10Why not n times unicast?
- Conferencing for residential users
- access asymmetric
- gt upstream bandwidth is limited
- gt point-to-multipoint mechanism required
- gt xcast is cheap multicast solution
- Examples xDSL, GPRS, cable modem, etc...
Access e.g. xDSL
Sender 1
Access router
Sender 2
2 Mb/s
200 Kb/s
11Earnest urgent need from a real user
- An anonymous PC-game vendor shipped a multiplayer
game using unicast UDP to share player status
information. - They considered using multicast but had concerns
about the number of concurrent - groups that could be supported in the existing
multicast schemes. - Currently they have 3000 groups but this number
is growing every day. And the reflector is
becoming a bottleneck.
If current multicast schemes are used, the number
of multicast groups in the route table will soon
surpass the number of unicast entries in the
route tables of backbone routers.
12Partnership
ISP
IT vendor
IRI (JPIX)
ITJIT
IBM
Fujitsu
Research Partners iCAIR/Internet2 W
IDE IIJ Lab SONY CSL EPFL
Alcatel
Communication Equipment vendor
13Xcast Partners
- iCAIR/Internet2
- WIDE Japanese research consortium
- IIJ Lab research subsidiary of www.iij.ad.jp
(active in IPv6 community). - SONY CSL is a SONY Center of Excellence.
- IRI (Internet Research Institute) one of the
most successful Internet ventures in Japan
(operates JPIX, Japans largest Internet
Exchange). - ITJITsubsidiary company of Nihon Telecom Co.
(Long Distance Carrier in Japan), serving mainly
domestic Internet connections. - Swiss Federal Institute of Technology / Ecole
Polytechnique Federale de Lausanne (EPFL)
14MDO6 Prototype
Router3
MDO6
MDO6
PEN4
PEN1
Router2
Router1
MDO6
PEN3
PEN2
- Based on FreeBSD 2.2.8 KAME
- KAME is a IPv6 stack developed by the WIDE
project - vic for IPv6 has been ported to MDO6
15SGM Prototype
- Host piece
- runs on AIX or Windows
- SGM library that
- implements the SGM layer/SGM API
- uses UDP and raw sockets to receivesend SGM
packets - a simple application
- Router stub
- an app that runs on AIX (or Windows)
- rolling out on RS/6000s used in the Internet2 DSI
project
16CLM Prototype
clm-vic
bottleneck
FreeBSD PC
17To Dos
- Develop Single Standard
- Work through some unresolved issues
- performance (several ideas exist to enhance
forwarding) - gradual deployment (several schemes exist)
- use of bitmap
- header format (L3 or L3.5)
- etc
18Summary
- Multicast scheme that scales with number of
sessions - Complementary to current multicast model
- Simple
- Several groups are working on similar approaches
- "Connectionless Multicast", draft-ooms-cl-multicas
t-01.txt - "Small Group Multicast", draft-boivie-sgm-00.txt
- "Multiple Destination option on IPv6",
draft-imai-mdo6-01.txt - "Distributed Core Multicast", draft-blazevic-dcm-m
obility-00.txt
19Next steps
- We submitted a proposal for a WG charter
(http//www.alcatel.com/xcast) - Presented xcast at MADDOGS meeting
- maybe in Internet Area
- main questions were on UDP checksum and IPsec gt
couple of scenarios possible - Meeting with Area Directors of Routing and
Internet areas
20Proposed Charter
- The Explicit Multicast (xcast) Working Group is
chartered to define, develop, specify and promote - principles and mechanisms that allow multi-point
to multi-point communication that scales with the - number of sessions.
- The basic idea of xcast is to use an explicit
list of destinations instead of one logical
multicast address. - This approach should be complementary to the
current multicast schemes. While today's
multicast - schemes are scaleable in the sense that they can
support very large multicast groups, these
schemes - are pretty expensive when a network needs to
support a very large number of distinct multicast
groups. - Explicit Multicast complements the existing
multicast schemes in that it can efficiently
support very - large numbers of distinct (small) multicast
groups and thus can play an important role in
making - multicast practical for applications such as IP
telephony, various conferencing collaborative - applications, multiparty networked games, etc
- The WG will try to reuse existing control plane
mechanisms and thus the primary focus of the WG
will be on data plane mechanisms. Initially an
end-to-end solution will be developed. Approaches
for both IPv4 and IPv6 will be defined. The WG
will specify the necessary protocols or protocol
extensions.
21Goals Milestones
- July 2000 Hold first Working Group meeting,
explain the "charter" and discuss the "Framework
Document". - Sept 2000 Submit "Framework Document" to IESG
(Informational RFC) - Feb 2001 Submit "Architecture Document" to IESG
(Informational RFC) - May 2001 Submit "Protocol Specification for IPv4"
to IESG (Standards Track) - May 2001 Submit "Protocol Specification for IPv6"
to IESG (Standards Track) - May 2001 Submit "MIB" to IESG (if needed)