Transcript WDM Multicasting via Optical Burst / Label Switching

WDM Multicasting via Optical
Burst / Label Switching
By
Krishna Kishore Konakanchi
Fall 2001 10/23/01
Overview
• Introduction
• Protocols requiring global knowledge of WDM
layer topology only
• Algorithms to construct a M-cast Forest given
global knowledge of both WDM layer topology
and multicast capability of the switches
• Re-routing protocol without global topology
as well as multicast capability info
• Future work
Introduction
• Supporting IP Multicasting over WDM networks has many
advantages including data rate and coding transparency for
multicast
• Main issue - multicast in WDM without changing the
semantics of IP
• All switches in WDM network are not MC i.e some are MI
• MC switch - can switch one incoming path onto many
downstream
• MI switch - Incoming signal can be passed through to only
one downstream
Layers 0-3 Characteristics (IP,ATM,SONET,WDM)
Network
Factors
IP
ATM
SONET
WDM
Data Protocol
Unit(PDU)
Varied –packet
size
53 byte cell
STS-1(45
Mbps)
Format
Independent 
Channel BW
Varied(logic)
Varied(logic)
Fixed
(Physical)
Fixed
(Physical)
Protocol Layer
Network
(layer 3)
Link
(layer 2)
Physical
(layer 1)
Optical
(layer 0)
Connection
Type
Connection
less
Connection
Oriented
Connection
Oriented
Connection
Oriented
QOS
Possible
Yes
NO
Possible
Traditional
Role
Routing
Switching
Cross
Connects
Cross
Connects*
Technology
Strength
Simple,easy scale,robust
Multi service
Integration
High speed
Transport
High – speed
Transport
* WDM
Emerging Technology
Protocols with global WDM
topology Knowledge
• Given a shortest path m-cast tree constructed by
MOSPF these protocols form a m-cast forest
avoiding branching at a MI nodes if present
• Protocol 1
– Re-route to source : MI switch having more than 1
downstream sends request towards source to find
alternate path. MC switch along the path acknowledges
by establishing new LSP
– no new modification to the SPT
– no hiding of routes from IP layer
– WDM need not know detail functionality of IP layer
Protocols with global WDM
topology Knowledge (cont.)
• Protocol 2
– Re-route to Any : determines paths to nodes on the tree
other than any of its children whose costs are less than
that of the path to source. Worst case is path to source
– Results in BW savings without substantial increase in
cost
• Consider following scenario where
– Source node : 1
– MI Node : 2
– Destinations : 3, 4 & 5
Shortest Path by MOSPF
1
2
4
3
5
Protocols 1 & 2 - Scenarios
Req.
&
Ack
1
2
4
1
3
5
2
4
3
Req.
&
Ack
5
Protocol 1
Protocol 2
Implementation & Results
• Protocols 1 & 2 were implemented in C &
OPNET
• Additional cost of forest constructed by 2nd rerouting protocol when compared with original cost
of tree is around 5% in a network with 20 - 40
switches , 50% of which are MI and 50% belong
to a session
• 2nd protocol can cut down cost by half when
compared to the 1st protocol.
Protocols with global WDM
topology Knowledge and MC Info
• Approach 1
– Construct forest based on any multicast tree
– At each branching MI switch, remove all
downstream links except one, thus breaking the
tree into base sub-tree and several superNodes
– Reconnect the base sub-tree with the closest
possible super node without using any removed
link
Protocols with global WDM
topology Knowledge and MC Info
(cont.)
• When its not possible to combine the base sub-tree
with any supernode , removed links are used to
form new LSP’s
• Resulting forest will consist of multiple source
routed trees.
• This is very similar to Re-route to Source.
KMB Tree heuristic to construct
Mcast Tree
• Given a graph “G” (modeling a network) and set
of multicast destination say “Z”
• The algorithm to construct the m-cast tree as
follows :
– Construct a complete directed distance graph
G1 = (V1,E1,c1)
– Find the minimum spanning tree T1 of G1
– Construct a sub-graph Gs of G by replacing each edge
in T1 by its corresponding shortest path in G
KMB Tree heuristic to construct
Mcast Tree (cont.)
– Find the minimum spanning tree Ts of Gs
– construct a Steiner Tree TH from TS by deleting edges in
TS if necessary, so that all the leaves in TH are Steiner
points.
• Worst Case Time complexity O(|S||V|2)
Approach 1 - Example
A
B
C
D
E
G
F
H
Approach 1 (cont.)
• From previous fig., the cost of the M-cast forest
can be given as
2(AB + BC + CD) + DE + EG + EH + DF
• This approach is very relevant in cases where the
higher layer is oblivious to the MI-nature of some
of the nodes. E.g IP over WDM
Protocols with global WDM
topology Knowledge and MC Info
(cont.)
• Approach 2
– Every member switch including the source acts as a
super node
– repeatedly combine the super-nodes into one until only
one super node containing all members are left
– when finding shortest path bet. 2 super nodes, only
paths without MI nodes are considered
– the downstream of a branching MI node may forward
multicast data to other down-streams thought an
OB/LSP
Approach 2 - Example
A
B
C
D
E
F
E
G
H
F
Approach 2 (cont.)
• The cost with this approach is given by
AB + BC + CD + 2(DE) + EG +EH+DF
• Gives the near optimal solution to the problem of
multicasting in a network where some/many nodes
do not have the
• It has been found that cost of forest using alg. 2 is
much lower than the 1st
Distributed Re-routing protocol with
neither Global Topology Nor MC
Info
• Basic idea - local multicast forwarding cache
• A branching MI switch sends a “purge” message
to all but one downstream switch
• Each purged switch then floods a “grow” request
to its neighbors
– direct grow : only neighbors already on the forest can
reply to the request
– indirect grow : if a neighbor is not on the forest then it
can still relay the request
Advantages of this Protocol
• Uses only local information of WDM layer
• Does not require any change to the IP multicasting
protocol
• Uses DVMRP as the IP multicast protocol, which
is most widely used in the internet
• Even if a MI switch is purged out, it can very
easily grow back into the forest with the grow
schemes
Distributed Protocol - Repair
Message
Repair
Repair
MC
MI
Repair
Repair
Repair
Repair
Purge
Purge
MC = Multicast-capable switch
MI = Multicast-incapable switch
Repair Messages
Distributed Protocol - Purge
Message
Purge
Purge
MC
MI
Gro
w
Repair
X
Repair
Y
Gro
w
Repair
Purge
Repair
Z
X
Z
Y
MC = Multicast-capable switch
MI = Multicast-incapable switch
Purge Messages
Purge
Repair & Purge
• MI can use randomly some heuristics to select down stream switch to
send repair messages
• Purged switch without any attached member need not have to grow
back to reduce signaling.
Grow Scheme
•Direct Grow Scheme
•Indirect Grow Scheme
•Neighbor not on multicast tree may replay grow request
Work to be done…
• The results obtained from the different methods
discussed are not comparable due to mismatch of
parameters used. They need to be generalized and
have to be run with same parameters to compare
the cost arrived at.
• The distributed protocol developed needs to be
implemented and its performance evaluated in
terms of BW and latency reduction.
Work to be done…(cont.)
• These results must be compared with the
results obtained from the other approaches
• New heuristics for supporting MOSPF,
protocols for PIM-SM and mechanisms to
provide QoS in multicasting can also be
given an insight.