MPLS and Traffic Engineering

Download Report

Transcript MPLS and Traffic Engineering

Week 9
MPLS: Multiprotocol Label
Switching
1
MPLS…What is it?
 MPLS – Multi-protocol Label Switching

can be applied for any layer 2 network protocol
 MPLS is a natural evolution of Internet
 It
is based on IP and routing protocols such as
BGP-4,OSPF and IS-IS
 Typically MPLS resides in service providers
networks and not in private networks
3
MPLS Protocol Stack
Application
IP or Multi-Service
MPLS
Layer 2 (PPP, ATM, FR,..)
Physical
4
What is MPLS ?
 MPLS provides connection oriented
switching based on a label applied at the
edge of an MPLS domain.
 IP is used to signal MPLS connections.
 Major Applications are:

Network Scalability

Traffic Engineering

VPNs
5
MPLS - Best Of Both Worlds
PACKET
Forwarding
HYBRID
IP
MPLS
1.
2.
3.
4.
5.
Scalable
Flexible
Easily Deployable
Inexpensive
Dynamic Routing
Packet
SWITCHING
ATM/FR
1.
2.
3.
4.
5.
Performance
Connection Oriented
Traffic Engineering
Security
QoS
Route at the Edge & Switch at the Core
6
Why MPLS ?
 MPLS converts connectionless IP to a connection-
oriented mode
 MPLS allows IP to be switched through the
Internet instead of routed
 With MPLS the first IP packet of a stream
establishes a switched path for all subsequent
packets to follow
 The packets will only be switched at each hop and
not routed.
7
Towards a connection-oriented
IP..
 MPLS is the evolution of current IP and connection
oriented protocols

Strength and scalability of IP routing

PVC like connectivity

ATM like QoS

Explicit routing
Plus

Path protection

Path optimization
8
MPLS Routers
 LER: Label Edge Router


Ingress LER examines inbound packets, classifies
packets, adds MPLS header and assigns initial label.
Egress LER removes the MPLS header and routes
packets as pure IP
 LSR: Label Switch Router

Transit switch that forwards packets based on MPLS
labels
9
IP and MPLS Forwarding
 IP packets are classified into FECs (Forwarding
Equivalence Class) at each hop based on the destination
address in conventional routing
 IP packet forwarding works by
 assigning a packet to a FEC
 determining the next-hop of each FEC
 MPLS group Packets by assigning a label in to FEC based
on Class of Service (CoS) and forward fast in a defined path
with the same forwarding treatment
10
Traditional IP Forwarding
Dest
47.1
47.2
47.3
Dest
47.1
47.2
47.3
Out
1
2
3
1 47.1
1
Dest
47.1
47.2
47.3
Out
1
2
3
IP 47.1.1.1
2
IP 47.1.1.1
3
Out
1
2
3
2
IP 47.1.1.1
1
47.2
47.3 3
2
IP 47.1.1.1
11
Label Switch Path (LSP)
 LSP: Label Switched Path
– Simplex L2 tunnel
– Equivalent of Virtual Circuit
 4 Byte Label is inserted in to the IP Packet at egress
MPLS node to Switch IP alone the created Label Switch
Path (LSP)
 Such MPLS nodes are called Label Switch Routers (LSR).
 With Labels IP Packet header is analyzed only at ingress
and egress LSRs where Labels are inserted and removed.
 Labels only have a Local significance and change from hop
to hop on the MPLS network. (Like DLCI in FR and
VCI/VPI in ATM)
12
Label Swapping
 Label Push: When an IP Packet enters the MPLS
network the Label is inserted by ingress LSR
 Label Pop: When the IP Packet exits the MPLS
network the Label is removed by the egress
LSR
 The Label is swapped at each intermediate hop
based on a Label mapping table
 Label mapping table is called the Label
Information Base (LIB)
13
MPLS Header
 Label: Label value, 20 bits

Exp: Experimental (CoS), 3 bits

ToS /DSCP to Exp mapping
 S: Bottom of stack, 1 bit
 TTL: Time to Live, 8 bit
– Ingress LER sets MPLS TTL to IP TTL
– Egress LER may set IP TTL to MPLS TTL or not
14
MPLS Operation
IP Forwarding
LABEL SWITCHING
Standard Routing
protocols
Labels are
exchanged . Egress
Ingress
LSR
Ingress LSR receives
IP packets, performs
packet classification
(into FECs), assigns a label, &
forwards the labeled packet
LSR
IP Forwarding
LSR
LSR
Egress LSR
LSRs forward
removes label before
packets based on
forwarding IP packets
the label (no
outside MPLS
packet classification
network
in the core)
Label
IP Hdr
Payload +
15
MPLS Packet Flow
Step 1: Ingress LER classifies IP packet , adds MPLS header and assigns label
Step 2: Transit LSR forwards label packet using label swapping
Step 3: Egress LER removes MPLS header and performs IP processing
16
MPLS IP forwarding via LSP
Intf Label Dest Intf Label
In In
Out Out
3
0.50 47.1 1
0.40
3
1
47.3 3
Label Dest Intf
In
Out
0.40 47.1 1
IP 47.1.1.1
1 47.1
Intf Dest Intf Label
In
Out Out 3
3
47.1 1
0.50
1
Intf
In
3
2
2
47.2
2
IP 47.1.1.1
17
MPLS Terminology
 LDP: Label Distribution Protocol
 LSP: Label Switched Path
 FEC: Forwarding Equivalence Class
 LSR: Label Switching Router
 LER: Label Edge Router (Useful term not in
standards)
18
Forwarding Equivalence
Classes LSR
LSR
LER
LER
LSP
IP1
IP1
IP1
#L1
IP1
#L2
IP1
#L3
IP2
#L1
IP2
#L2
IP2
#L3
IP2
IP2
Packets are destined for different address prefixes, but can be
mapped to common path
• FEC = “A subset of packets that are all treated the same way by a router”
• The concept of FECs provides for a great deal of flexibility and scalability
• In conventional routing, a packet is assigned to a FEC at each hop (i.e. L3
look-up), in MPLS it is only done once at the network ingress
19
LABEL SWITCHED PATH
(vanilla)
#216
#14
#311
#99
#311
#963
#311
#963
#14
#612
#5
#462
#99
#311
- A Vanilla LSP is actually part of a tree from
every source to that destination (unidirectional).
- Vanilla LDP builds that tree using existing IP
forwarding tables to route the control messages.
20
MPLS BUILT ON
STANDARD IP
Dest
47.1
47.2
47.3
Out
1
2
3
Dest
47.1
47.2
47.3
Out
1
2
3
1 47.1
3
1
Dest
47.1
47.2
47.3
Out
1
2
3
2
3
2
1
47.2
47.3 3
2
• Destination based forwarding tables as built by OSPF, IS-IS, RIP, etc.
21
IP FORWARDING USED BY HOP-BYHOP CONTROL
Dest
47.1
47.2
47.3
Dest
47.1
47.2
47.3
Out
1
2
3
1 47.1
1
Dest
47.1
47.2
47.3
Out
1
2
3
IP 47.1.1.1
2
IP 47.1.1.1
3
Out
1
2
3
2
IP 47.1.1.1
1
47.2
47.3 3
2
IP 47.1.1.1
22
MPLS Label Distribution
Intf Label Dest Intf Label
In In
Out Out
3
0.50 47.1 1
0.40
Intf
In
3
Label Dest Intf
In
Out
0.40 47.1 1
1
Request: 47.1
3
Intf Dest Intf Label
In
Out Out
3
47.1 1
0.50
3
2
1
1
47.3 3
47.1
Mapping: 0.40
2
47.2
2
23
Label Switched Path (LSP)
Intf Label Dest Intf Label
In In
Out Out
3
0.50 47.1 1
0.40
Intf Dest Intf Label
In
Out Out
3
47.1 1
0.50
3
1
47.3 3
Label Dest Intf
In
Out
0.40 47.1 1
IP 47.1.1.1
1 47.1
3
1
Intf
In
3
2
2
47.2
2
IP 47.1.1.1
24
Label Distribution Protocols
 The network automatically builds the routing tables by
IGP protocols such as OSPF, IS-IS. The label
distribution protocol (LDP) uses the established routing
topology to route the LSP request path between
adjacent LSRs.
 A LDP is a set of procedures by which one LSR informs
another LSR of the label to FEC bindings it has made.
 The LDP also encompasses any negotiations in which two
label distribution peers (Ingress LSR and egress LSR)
need to engage in order to offer CoS to particular IP
stream.
25
Hop-by-hop routed LSP
0
A
1
2
0
1
E
0
2
1
B
2
1
2
0
D
C
192.6/16
0
Incoming
Label
Outgoing
Label
Next hop
Outgoing
Interface
A
100
?
B
1
B
6
?
E
1
C
17
?
D
2
D
5
?
E
0
E
6
?
E
0
26
Example Contd
Incoming
Label
Outgoing
Label
Next hop
Outgoing
Interface
A
100
6
B
1
B
6
6
E
1
C
17
5
D
2
D
5
6
E
0
E
6
?
E
0
27
EXPLICITLY ROUTED
OR ER-LSP
Route=
{A,B,C}
#14
#972
#216
B
#14
A
C
#972
#462
- ER-LSP follows route that source chooses. In
other words, the control message to establish
the LSP (label request) is source routed.
28
EXPLICITLY ROUTED LSP
ER-LSP
Intf Label Dest Intf Label
In In
Out Out
3
0.50 47.1 1
0.40
Intf
In
3
3
Dest
47.1.1
47.1
Intf
Out
2
1
Label
Out
1.33
0.50
3
1
47.3 3
Label Dest Intf
In
Out
0.40 47.1 1
IP 47.1.1.1
1 47.1
3
1
Intf
In
3
2
2
47.2
2
IP 47.1.1.1
29
ER LSP - advantages
•Operator has routing flexibility (policy-based,
QoS-based)
•Can use routes other than shortest path
•Can compute routes based on constraints in exactly
the same manner as ATM based on
distributed topology database.
(traffic engineering)
30
Routing Scalability via MPLS
Routing
Domain C
Routing
Domain B
X
V
T
W
Z
Y
Routing Domain A
31
Routing Domain A: Routes to W
Incoming
Label
Outgoing
Label
Next hop
T
N/A
10
X
X
10
12
Y
Y
12
17
W
W
17
N/A
W
32
Example Contd – Label Stacking
Routing
Domain C
Routing
Domain B
V
10
2
5
T
X
12
2
6
17 W
2
Z
Y
Routing Domain A
33
Routing Transients
R1
R5
R4
R2
R3
 Routing transients happen due to



failure detection (in the order of milliseconds)
LSP dissemination (in the order of propagation
delays)
SPF tree calculation (in the order of several
hundred milliseconds)
34
MPLS Example
LSR9
LSR8
LSR4
LSR2
Pop
LSR1
14
37
LSR5
LSR6
LSR7
Setup: PATH (ERO = LSR1, LSR2, LSR4, LSR9)
Labels established on RESV message
35
Fast Reroute - Protection Path
LSR8
LSR9
LSR4
LSR2
Pop
LSR1
17
LSR5
LSR7
LSR6
22
Setup: PATH (LSR2, LSR6, LSR7, LSR4)
Labels established on RESV message
36
LSR8
Example
Swap 37 --> 14
Push 17
Pop 14
LSR4
LSR9
LSR2
Push 37
LSR1
LSR5
LSR6
Swap 17 --> 22
LSR7
Pop 22
Label Stack LSR1 LSR2 LSR6 LSR7 LSR4
37
17
22
14 None
14
14
37
MPLS Protection
 May result in suboptimal forwarding but service
interruption is negligible
 A single protection LSP could be used to fast-
route not one but multiple LSPs
 Protection on a per-LSP basis (end-to-end) rather
than on a per-link basis is also possible

better forwarding properties in case of failures

a single protection may not protect as many LSPs

handles both node and link failures

detection time may be larger

will require the computation of link and node disjoint
paths
38
Label Encapsulation
L2
ATM
FR
Label VPI VCI DLCI
Ethern PPP
et
“Shim Label”
“Shim Label” …….
IP | PAYLOAD
MPLS Encapsulation is specified over various
media types. Top labels may use existing
format, lower label(s) use a new “shim” label
format.
39
Traffic Engineering - Objectives
 Performance optimization of operational
networks.

Reduce congestion hot spots.

Improve resource utilization.
 Why current IP routing is not sufficient from
TE perspective? Fish problem.

Destination-based

Local optimization
40
IP Routing & “the Fish”
R8
R3
R4
R2
R5
R1
R6
R7
IP (Mostly) Uses Destination-Based Least-Cost Routing
Flows from R8 and R1 Merge at R2 and Become Indistinguishable
From R2, Traffic to R3, R4, R5 Use Upper Route
Alternate Path Under-Utilized
6
Deficiencies in IP Routing
 Chronic local congestion
 Load balancing

Across long haul links
 Size of links

Difficult to get IP to make good use unequal
size links without overloading the lower speed
link
42
Peer Model
 Peer model

OSPF routing + link weights.

Key technique: weight setting.

Networks operate as it does today.

Much more scalable than overlay model.
43
Load Balancing
Making good use of expensive links simply by
adjusting IGP metrics can be a frustrating exercise!
44
Overlay Motivation
Separate Layer 2 Network
(Frame Relay or ATM)
“The use of the explicit Layer 2 transit
layer gives us very exacting control of
how traffic uses the available bandwidth
in ways not currently possible by
tinkering with Layer 3-only metrics.”
The Overlay Solution
L3
L3
L2
L3
L2
L2
L2
L3
L2
L3
L2
L3
L3
L3
L3
L3
L3
L3
Physical
Logical
Layer 2 (for example ATM) network used to
manage the bandwidth
Layer 3 sees a complete mesh
Overlay Drawbacks
Extra network devices (cost)
More complex network management
Two-level
network without integrated NM
Additional
training, technical support,
field engineering
IGP routing doesn’t scale for meshes
Number
of LSPs generated for a failed router is
O(n3); n = number of routers
Overlay Drawbacks
 Every router is permanently connected to every other
router (fullmesh)



PVCs are provisioned with given bandwidths
Delays are short
Problem: scalability
• For N routers, N x (N-1)/2 ATM VCs
• Also:
• The IP link-state routing protocol (e.g. OSPF) has to handle a huge
number of links, and link State Advertisements packets are flooded
on every link

Worse: when an ATM link fails, all VCs using that link fail,
andmany IP routers have to update their routing tables at the
same time
 Amount of routing information can be as much as N^4
 In practice, this solution does not scale beyond 100 routers
(± 5000 PVCs)
48
Traffic Engineering & MPLS
+
Router
ATM Switch
or
=
MPLS
Router
ATM MPLS
Router
MPLS fuses Layer 2 and Layer 3
Layer 2 capabilities of MPLS can
be exploited for IP traffic engineering
Single box / network solution
An LSP Tunnel
R8
R3
R4
R2
R5
R1
R6
R7
Labels, like VCIs can be used to establish virtual circuits
Normal Route R1->R2->R3->R4->R5
Tunnel: R1->R2->R6->R7->R4
50
Comprehensive Traffic
Engineering
 Network design

Engineer the topology to fit the traffic
 Traffic engineering
 Engineer

the traffic to fit the topology
Given a fixed topology and a traffic matrix,
what set of explicit routes offers the best
overall network performance?
51
Constraint-based routing
 Two basic elements


Route optimization: Select routes for traffic demands subject to
a given set of constraints.
Route placement: Implement the selected routes in the network
so that the traffic flows will follow them.
 Mathematical formulation
 Assumptions


Network is represented as a directed graph G(V, E).
Network links and capacities are directional.

Average traffic demand is known.

Traffic demand between two edge nodes is directional.
 Objectives

All traffic demands are fulfilled.

Minimize the maximum of link utilization.
52
Off-line Formulation
 Notations
 G=(V,E)
 cij be the capacity of link (i,j), for all (i,j) in E.
 K: the set of traffic demands between a pair of
edge nodes.
 (dk,sk,tk): (bandwidth demand, source node,
destination node), for all k in K.
 Xijk the percentage of k’s bandwidth demand
satisfied by link (i,j).
 α: the maximum of link utilization among all the
links.
53
Off-line Formulation
54
On-line TE
 Shortest path (SP)

Link metric for link (i,j) is inversely proportional
to the bandwidth.
 Minimize the total resource consumption
per route.

Minimum hop (MH)

Link metric is set to 1 uniformly for each hop.

Still run shortest path algorithm.
55
On-line TE
 Shortest-widest path (SWP)



Link metric is set to as bandwidth.
Always selecting the path with largest bottleneck
bandwidth.
The one with minimum hops or shortest distance is
chosen when multiple paths are available.
 Hybrid algorithm



Motivation.
Solution: appropriate weight assignment, and link
utilization (instead of using link residual bandwidth).
Metric: path cost + link utilization.
56
Mathematical Formulation
 Notations
 fij: current load (used capacity) of link (i,j); initial value is 0
 cij: total capacity of link (i,j)
 α: current maximum link utilization; initial value is 0
 αij: link (i,j) cost metric
57
Multi-path Algorithms
L1
Incoming
Traffic
Traffic
Splitter
Outgoing
Traffic
L2
58
Traffic Splitting
 Basic requirements


Traffic splitting is in the packet-forwarding path, and
executed for every packet.
To reduce implementation complexity, the system should
preferably keep no or little state info.
 Traffic-splitting schemes produce stable traffic
distribution across multiple outgoing links with
minimum fluctuation.
 Traffic-splitting algorithms must maintain per-
flow packet ordering.
59
Hashing
 Direct hashing

Hashing of destination address

H(•)=DestIP mod N

N: the number of outgoing links
 Hashing using XOR folding of source/destination addresses
 H(•)=(S1⊗S2⊗S3⊗S4⊗D1⊗D2⊗D3⊗D4) mod N

⊗: XOR operation

Si: the ith octet of the source address

Di: the ith octet of the destination address

CRC 16 (16-bit cyclic redundant checksum)
 H(•)=CRC16(5-tuple) mod N
 (−) distributing traffic load evenly
60
Hashing
 Split a traffic stream into
M bins.
 The M bins are mapped to
N outgoing links based on
an allocation table, i.e.,
compute the corresponding
hash value.
 By changing the allocation
of the bins to the outgoing
links, we can distribute
traffic in a predefined
ratio.
1
N
2
1
3
1
4
N
1
N
M-1
3
M
1
61
Bilkent’s Traffic Engineering
Onur Alparslan’s MS thesis work
• Introduction
•
Traffic Engineering (TE)
•
Problem Statement
• Our Proposed TE Architecture
•
Path Establishment
•
Queuing Models
•
Feedback Mechanism and Rate Control
•
Traffic Splitting Algorithm
• Simulations
• Conclusion
62
Traffic Engineering (TE)
Definition: The process of controlling how traffic flows through a
network so as to optimize resource utilization and network
performance, and reconfiguration of mapping in changing network
conditions.
Advantages:
• Provide ISPs precise control over the placement of traffic flows.
• Balance the traffic load on the various links, routers, and
switches in the network so that none of these components is
overutilized or underutilized.
• Provide more efficient use of available aggregate bandwidth.
• Avoid hot spots in the network.
63
Problem Statement
Our Goal:
• Our main aim is to increase total amount of carried
traffic and balance the load of links in the network by
using two disjoint paths (multipath).
• No need for prior information on traffic matrix.
• Eliminate knock-on effect.
• Consider the load balancing performance of elastic
TCP flows.
• Apply methods that are TCP friendly
• Capabilities to simulate a mesh network with thousands
of TCP flows by using ns-2
64
Problem Statement
Our Approach:
• A primary and a disjoint secondary path are
established from an ingress node to each egress node.
• Split TCP traffic between the primary and secondary
paths using a distributed mechanism based on ECN
marking and AIMD-based rate control.
• Primary paths have strict priority over the secondary
paths with respect to packet forwarding
• TCP splitting mechanism operates on a per-flow basis
in order to prevent packet reordering which can
substantially reduce TCP performance
65
Path Establishment Without Traffic Information
• We establish two disjoint paths between each source destination
pair
• The first one is the Primary Path (PP) and uses shortest path
found using Dijkstra’s algorithm.
• The second one is the Secondary Path (SP) and it is computed
after pruning the links used by PP and using Dijkstra’s algorithm in
the remaining network graph.
Node 3
Node 1
Destination
Node 2
Node 5
Source
Node 4
66
Queuing Model
Egress Node 1
Backbone Network
Per-egress queuing
PP Queue
SP Queue
PP Queue
Per-class queuing
Gold Queue
RM + TCP ACK
Silver Queue
SP Queue
Bronze Queue
Egress Node 0
Egress Node 2
67
Knock-on Effect
• Giving equal priority to PPs and SPs may decrease the
performance of PPs since a SP may share links with PPs of
other node pairs
• Traffic increase on a SP may force sources of PPs sharing
links with this SP to move traffic to their own SPs
• This further decreases performance, because SPs typically
use longer routes and can also force other PPs to move traffic
to their SPs
Edge 3
Edge 1
Edge 2
68
Bistability in Single Overlay: Phone
Network
 Phone network is an overlay

Logical link between each pair of switches

Phone call put on one-hop path, when possible

… and two-hop alternate path otherwise
 Problem: inefficient path assignment

Two-hop path for one phone call

… stops another call from using direct path

… forcing the use of a two-hop alternate path
busy
busy
69
Preventing Inefficient Routes:
Trunk Reservation
 Two stable states for the system

Mostly one-hop calls with low blocking rate

Mostly two-hop calls with high blocking rate
 Making the system stable

Reserve a portion of each link for direct calls

When link load exceeds threshold…
• … disallow two-hop paths from using the link

Rejects some two-hop calls
• … to keep some spare capacity for future one-hop calls
 Stability through trunk reservation

Single efficient, stable state with right threshold
70
FIFO (First In First Out) Queuing
•
TCP data packets of PPs and SPs join the same silver queue and we
do not make use of the Bronze Queue at all
•
ACK and Probe Packets (RM) join the Gold Queue
•
Gold Queue has strict priority over Silver Queue.
Gold Silver
ACK
Probe Packet (RM)
PP Data Packet
SP Data Packet
Core Router
71
SP (Strict Priority) Queuing
•
Data packets of PPs enter Silver Queue. Data packets of SPs enter
Bronze Queue
•
ACK and RM Packets join the Gold Queue
•
Gold Queue has strict priority over other queues, Silver Queue has
strict priority over Bronze Queue
Gold Silver Bronze
ACK
Probe Packet (RM)
PP Data Packet
SP Data Packet
Core Router
72
Hybrid SP – Deficit Round Robin
Scheduler
 Give priority to TCP ACK and RM packets
 Of the remaining capacity

90 % is given to PP flows
 10
% is given to LP flows
 Very similar to strict priority queueing
except that SP flows are not starved
73
Feedback Mechanism
Gold Silver Bronze
ACK
Primary RM (P-RM)
Secondary RM (S-RM)
Primary Path Data Packet
0
0
0
0
Secondary Path Data Packet
Core
0
0
0
0
Gold Silver Bronze
Ingress
Egress
1
0
0
1
0
1
Core
0
0
1
74
Feedback Mechanism
ACK
Primary RM (P-RM)
Secondary RM (S-RM)
Primary Path Data Packet
0
0
0
0
Core
0
Ingress
0
Secondary Path Data Packet
0
0
0
Egress
1
1
1
1
1
1
Core
1
1
1
75
Rate Control
• When the Ingress Node receives the congestion information
about the path, it will invoke an AIMD (Additive Increase
Multiplicative Decrease) algorithm to compute the ATR
(Allowed Transmission Rate) of the corresponding path.
ATR: Allowed Transmission Rate
RDF: Rate Decrease Factor
RIF: Rate Increase Factor
MTR: Minimum Transmission Rate
PTR: Peak Transmission Rate
76
Traffic Splitting
Traffic Splitting Units
Per-egress queuing
DPP
Incoming Flows
For Destination 1
AIMD
PP Queue
Per-class queuing
Gold Queue
+
RM + TCP ACK
SP Queue
DSP
DPP
Incoming Flows
For Destination 2
Silver Queue
PP Queue
+
SP Queue
Bronze Queue
DSP
•
•
•
When a new flow arrives at an ingress router, a decision on how to
forward the packets of this new flow needs to be made.
We compute the DPP and DSP delay estimates for the PP and SP
queues at the Edge Node, respectively.
Then calculate and update dn that is averaged (smoothed)
difference, DPP - DSP, at the epoch of the nth packet arrival
77
Random Early Reroute
• By using the updated dn value, we decide whether to assign
this new flow to PP or SP:
•
Assign the new flow to PP with probability (1-p(dn))
•
Assign the new flow to SP with probability (p(dn))
• We call this policy as Random Early Reroute (RER). It is
used for controlling the delay difference of queues of PP
and SP
• This policy gives priority to PP over SP on the edge nodes.
78
Simulation Setting
Three Node Network Topology
Edge 2
•
Flow size dist. = Bounded Pareto
•
Flow interarrival dist. = Poisson
•
Total traffic from each node = 70 Mb/s
•
Speed of core links = 50 Mb/s
Core 2
Core 1
Edge 1
Core 3
Edge 3
79
Simulation Parameters
Mesh Network Topology
ch
cl
de
ny
sl
sf
dc
sj
at
da
la
hs
• This topology and the traffic demand matrix T[i,j] are used from the
data given in www.fictitious.org/omp.
• Each link is bi-directional and has 45 Mb/s capacity in both
directions except the links between de-ch and ch-cl have 90 Mb/s
capacity in both directions.
80
Simulation Setting
• TRM = 0.1 s for 3-node,
0.02s otherwise
• p0 = 1
• RER:
• minth= 1 ms
• maxth= 15 ms
• Shortest Delay:
• minth= 0 ms
• maxth= 0 ms
Proposed architecture is implemented over ns2 network simulator as an extension
81
Results - 3 nodes
83
Results – Large topology
84
Triangle Effect – 3 Nodes
85
Large Topology
86