View
224
Download
0
Category
Tags:
Preview:
Citation preview
Multicasting In MPLS Network
Christy Gnanapragasamchristy@sce.carleton.ca
Nov 08, 2003
July 25, 2003AONL & EION 2
Outline• Introduction and Background• Literature survey
– IP Multicasting – Multicasting in MPLS
• Difficulties in supporting• A New approach (ERM2)• Extension to ERM2 (ERM2-Ext) (Own
work)• Performance analysis• Summary
July 25, 2003AONL & EION 3
Introduction and Background
July 25, 2003AONL & EION 4
• Advantages of multicasting– Reduction in network resource consumption– Source Link stress
• Example applications:– Audio/video distribution– Push applications– Audio/video conferencing
• Requirements of this applications (QoS):– Bandwidth, bounded delay and low loss rate
Introduction
July 25, 2003AONL & EION 5
• There is no standardized algorithms for multicasting in MPLS
• However, the power of MPLS can provide QoS for IP multicast in MPLS.
• Problem arise, when mapping L3 multicast trees in to L2 LSPs.– Flood and prune– Source/shared trees– Uni/bi directional trees– Encapsulated multicast
Difficulties in supporting IP multicast in MPLS
July 25, 2003AONL & EION 6
ERM2• ERM:
– Easy to implement and it requires simple modification to current IP mulicasting protocols– But, LSRs need to participate in routing process
• ERM2:– In MPLS-TE, Until a PLR has a backup path available, the PLR must clear relevant four flags in the
corresponding RRO sub-object
– Local Protection available flag• PLR should set this when a backup path is available. • If no established backup path or it is in DOWN state, PLR should clear this flag and should send a updated RESV message
– Local Protection in use flag• PLR Should set this flag if it redirecting traffic into the backup path else should be clear
– Node Protection flag• PLR should set this flag if backup path is protected against the failure of immediate downstream node.
– BW Protection flag• PLR should set this flag if the backup path offers a BW guarantee.
July 25, 2003AONL & EION 7
• Difficulties:
– LSP Design • Multicast tree requires point-to-multipoint LSP setup, but only point-
to-point LSP setup is supported by MPLS. • Volatile LSPs
– Dynamic behavior of group members– Singling overhead and over consumed labels
– Traffic aggregation:• Unicast traffics are normally aggregated if ingress and egress LERs
are same for scalability. • Not possible in multicast traffic.
– Coexistence of Layer 2 and Layer 3 forwarding in LSRs.• Two cases:
– Switched over from shared tree to source based tree– Same label for unicast and multicast traffic
July 25, 2003AONL & EION 8
• Edge Router Multicasting
• In this approach, trees are formed by branching at Edge router.
Solution to mentioned problems
July 25, 2003AONL & EION 9
Advantages of ERM• ERM converts the multicast flow into multiple unicast
flow
• Simplifies the LSP setup: – Diverging nodes are only LERs– No need for point-to-multipoint LSPs.
• Makes multicast flow aggregatable:– Unicast and multicast traffic can be treaded in a same way and
scalability will not be compromised
• Relaxes the requirements at Core Routers:– No modification needed at LSRs
July 25, 2003AONL & EION 10
ERM in depth
• ERM concept can be used to build trees in MPLS by slightly modify existing IP multicast routing protocols.– PIM_SM and CBT uses Rendezvous Point
and explicit Join messages– But need some modifications needed
• Select LER as RP of the tree• Allow a sub tree to join at only LER
July 25, 2003AONL & EION 11
• We can add two more techniques that can optimize the end results.– Where to join if there are multiple choices.– Dynamic tree optimization
• Advantages.– Minimize the End-to-End delay– Minimize the bandwidth consumption– Minimize the packet processing (L2->L3->L2)
Modification to ERM2 (own work)
July 25, 2003AONL & EION 12
RESULT
LER1• ERM2 illustrations
LER6
LER5
LER4
LER3
LER2
MM
R2
QUARY
R1
JOIN
ADD
S
PRUNESUBTRACT
LER1
LER6
July 25, 2003AONL & EION 13
RESULT
LER1
• ERM2-Ext Joining choice
LER6
LER5
LER4
LER3
LER2
MM
R2
QUARY
R1
SLER1 3
LER5 0
July 25, 2003AONL & EION 14
LER1
• ERM2-Ext: Dynamic tree optimization
LER6
LER5
LER4
LER3
LER2
MM
R2
R1
S
1) Unwanted packet processing
(L2->L3->L2)
3) Extra bandwidth wastage (5 link -
>4link)
2) ETE hop delay is larger than it should
be (5->4)
July 25, 2003AONL & EION 15
• Challenges:
– Triggering the optimization algorithm• Optimization can be triggered when a LER does not have any directly
connected receivers and it has one or more downstream peer LERs.
– Optimization• Which nodes should be refreshed (rearranged)• Where should they join (who is the parent)• Which order those nodes should be rearranged
Introducing a simple algorithm would optimize the tree and the performance
July 25, 2003AONL & EION 16
Snapshot of a tree
R
All of these nodes are LERs and they
all have directly attached Receivers
5
4
5
3
3
5 6
3 45
5
3
2
Which nodes need to be refreshed:
children
Where should they join:
Easy solution: join them in parent node of the leaving LER
1) Which children to refresh first
2) Find the potential candidates (LERs)
MM
RefreshTrigger (list of children)
Refresh (potential candidates)
July 25, 2003AONL & EION 17
• Find a order: which children need to be refreshed first to obtain optimal result.– Chose the one has less cost to source (root)– This new child also be part of the potential candidate for next child.
• Finding the potential candidates– Which is a sub-set of on-tree LERs– Sub-tree should exclude
• Leaving LER• Children of leaving LER
– What about descendant of leaving node• Should be excluded
– Why?
MM responsibilities:
July 25, 2003AONL & EION 18
Find the potential candidates for joining the tree
R
5
4
5
3
3
5 6
3 45
5
3
2
MM
MM does not know the relationship between nodes, it just know the distance from
source
To exclude descendants, may be we can use the height for
now
July 25, 2003AONL & EION 19
Bandwidth consumption
Bandwidth consumption
0
20
40
60
80
100
120
140
1 12 23 34 45 56 67 78 89 100
Time
BW
link utilization(ERM2-EXT)
link Utilization(ERM2)
link utilization (ERM2-Ext_completed)
July 25, 2003AONL & EION 20
End to End DelayETE Delay (num of hops)
0
1
2
3
4
5
6
1 12 23 34 45 56 67 78 89 100
Time
nu
m h
op
s
ETE delayERM2_Ext(time-average)
ETE delay ERM2(time_average)
ETE delay_ERM2-Ext-completed(timeaverage)
July 25, 2003AONL & EION 21
Summary• ERM2 is a simple algorithm that supports the
multicasting in MPLS.• It excludes the need for point-to-multipoint LSP setup• It excludes the participation of LSR in tree or tree
building
• ERM2-Ext: for more optimization– Choice of parent when multiple joining points are possible– Dynamically refreshing the tree
• Two new messages– RefeshTriggring and Refresh
– Advantages• Minimize the bandwidth consumption• Minimize the end to end delay• Minimize the processing overhead (L2->L3->L2)
July 25, 2003AONL & EION 22
References• [1] Boudani, A and Cousin, B.;”A New Approach to Construct Multicast Trees in MPLS
Network,” Computers and Communications, 2002. Proceedings. ISCC 2002. Seventh International Symposium on, 1-4 July 2002 Pages: 913 – 919.
• Baijian Yang and Mohapatra, P.” Edge router multicasting with MPLS traffic engineering,” Networks, 2002. ICON 2002. 10th IEEE International Conference on, 27-30 Aug. 2002 Pages: 43 – 48
• Young-Kyu Oh, Dong-Keun Kim, Hun-Je Yoen, Mi-sun Do and Jaiyong Lee,” Scalable MPLS multicast using label aggregation in Internet broadcasting systems,” Telecommunications, 2003. ICT 2003. 10th International Conference on, Volume: 1, 23 Feb.-1 March 2003.Pages:273 - 280 Vol.1
• Zhongshan Zhang, Keping Long, Wendong Wang, Shiduan Cheng, “The new mechanism for MPLS supporting IP multicast,”, Circuits and Systems, 2000. IEEE APCCAS 2000. The 2000 IEEE Asia-Pacific Conference on, 4-6 Dec. 2000 Pages: 247 – 250
• M. Ramalho. Intra- and Inter-domain multicast routing protocols: A survey and taxonomy. IEEE Communications Survey sand Tutorials, 3(1):2–25, first Quarter 2000
• Almeroth, K.C., “The evolution of multicast: from the MBone to interdomain multicast to Internet2 deployment,“ Network, IEEE , Volume: 14 , Issue: 1 , Jan.-Feb. 2000, Pages:10 – 20
Thank You
July 25, 2003AONL & EION 24
Xcast solution
• Explicit multicast [Boivie & al.]
• Xcast packet – explicit list of group members – list of unicast addresses in Xcast packet header
• No multicast routing table– standard unicast routing protocol– no additionnal entry
• No management protocol
July 25, 2003AONL & EION 25
MPLS Multicast Tree
• uses LSPs between branching nodes
• no multicast states into no-branching nodes
• enhances scalability of MPLS domain for multicast trafic
July 25, 2003AONL & EION 26
Branching nodes
• Multicast trees have: – few branching nodes– many non-branching
nodes
• A branching node– a router with distinct next
hops for a multicast tree
from REUNITE study
from our study
July 25, 2003AONL & EION 27
Xcast problems
• Membership management– delegated to another protocol
• multicast source knows somehow the IP address of every member of the multicast group
• When size of multicast group increases – size of destinations list increases
• packet processing time increases• packet length increases
July 25, 2003AONL & EION 28
Multicasting in MPLSIn designing a MPLS multicast protocol, we must
deal with two problems.1. Creating multicast tree.2. Allocating and distributing labels.Definitely, LDP must deal with the second
problem. However, creating Multicast tree can be based on:– Topology driven: Map the L3 tree in the routing
protocol into L2.– Request driven: By signaling– Traffic driven: Map the L3 tree in the routing
protocol into L2 when data arrive.
July 25, 2003AONL & EION 29
Xcast packet forwarding
A B
C
D
E
F
d1
d3, G
d2, G
s1
d1 : E
dest : next hop
d4, G
d2 : Cd3 : Dd4 : D
G : d1,d2,d3
s1->G : Xcast d2, d3, d4 <data>
grp: list of dest
s1->G : Xcast d3, d4 <data>
s1->d2 <data>
- Lookup for every destinations into Xcast packet
- Forward a copy of the packet (with an appropriated destination list) for every distinct next hop
July 25, 2003AONL & EION 30
MPLS domain
Multicast tree over MPLS• In a MPLS domain, usually
– packets is forwarded over the LSP ending at the edge LSR associated to the packet destination
– paths followed by multicast packets will not be efficient
ingress LSR egress LSR
LSP
July 25, 2003AONL & EION 31
SEM principles
• Use of the next branching node address as Destination address of IP packet– no specific treatment into routers which are
between branching nodes
• SEM packet header encapsulates Multicast group address– processed by branching nodes– increasing of packet length is low
July 25, 2003AONL & EION 32
Segmentation for XCast
• Overhead of Xcast increases with the increasing of the number of destinations– best tradeoff between path-MTU length and
maximum number of destinations into a Xcast packet
– proposition of GXcast protocol: a generalization of Xcast
• partition of the set of destinations based on the optmized size of the parts
• how to find the best partition [still an open question]
July 25, 2003AONL & EION 33
July 25, 2003AONL & EION 34
July 25, 2003AONL & EION 35
IGMP is only used between hosts and first hop router; for router-to-router distribution, a routing protocol is used to build and maintain group distribution trees
July 25, 2003AONL & EION 36
July 25, 2003AONL & EION 37
MOSPF, which is basically the OSPF routing protocol with extensions, makes use of the SPF (Shortest Path First) algorithm. All the other multicast protocols can be split into two types: "broadcast and prune" and "explicit join." Broadcast and prune protocols (DVMRP, PIM-DM) typically build source-specific distribution trees (fig. A) while explicit join protocols (PIM-SM, CBT) build shared trees (fig. B) from a Rendezvous Point (RP).
Recommended