MPLS convergence –
IGP and BGP Convergence Impacts
-By Susan Hares
[email protected] Why worry about Performance:
Reduce Operational expense
Reduce downtime Convergence
Reduce outages due to
configuration and provisioning
errors
Reduce outages due to Security
attack
Increase services Memory footprint
VPN services
VOIP services Scaling
Access Services
2
Convergence issues
LDP L2 & L3 VPNs
IGP shortcuts BGP carries labels or
Forward Endpoint info (VPLS)
Adjacencies
BGP
Impacts Convergence
Link IGP
Failure Impacts Convergence interacts
Detect
interacts RSVP
Convergence
3
Caveats on Model
Model versus Reality
Believe the Real Data
Real IPv4 Data: Beacon Experiments, Route Views, Telstra
Data base
Problem: No L3 VPN data recorded separate
Model tries to match benchmarking with Real data
1st pass based on IGP + BGP models
1st pass – light on the LDP and RSVP models
Feedback on Model’s appreciated
4
MPLS Model of Convergence
Models for Convergence
Link, IGP, BGP, LDP, RSVP-
TE
Route Traffic models
Benchmark
Single Node vs Multiple
Nodes
Results
5
Link Failures Times
Link
Failure Model Equation: D = Minimum(3*Link-timer, S-
Detect out)
Best: Hardware detection
Planned: Link-Timer : “alive” timer for
the link
S-out = 3 times Link-Timer
Physical link
Sonet:Link Time up/down = 10 ms
Improvements if change to 1 ms
Ethernet, Wireless: Use of Signaled Link Down
IP layer - BFD light weight Hello
6
IGP Convergence times
LoC(p) = D + O + QSP + (h * F) + SPF(n) + RIB (p) + FIB(p)
+ DD + CRR
D: Failure is detected
O: New LSP is originated
Dominate
Insertion of External routes occurs here
variables
QSP: cumulative queuing, serialization, propagation, With high
H*F: LSP is flooded up to rerouting node speed/bandwidth
SPF: Shortest path calculation links
RIB: RIB is updated
FIB: FIB is updated References: www.nanog.org/mtg-
DD: LCs are updated 0202/ppt/cengiz.pdf; mtg-0010/ppt/cengiz.pdf
CRR: BGP recursion is fixed
7
IGP Topologies (OSPF, ISIS)
N R N N R N
R N R R N R
N R N N R N
Router = Router LSA Router = Router
N = Network LSA LSA
Heavily Meshed N = Network LSA
Medium Mesh
Types of topology
Lightly connected (1-2 links/ router)
Medium connected (2-3 links/router
Highly connected (4-5 links/router)
Least Squares Test topology
8
OSPF Convergence time
Convergence for LSA Square Network
350.00
300.00
Time (milliseconds)
250.00
200.00
150.00
100.00
50.00
0.00
13 50 313 1250 5000
Networks
OSPFv2 SPF (ms) OSPFv3 (ms) Cisco's Fast SPF (ms)
Caveat: Cisco utilized medium mesh; GateD utilizes heavily meshed
9
OSPF External Routes caveat
The deployment of the OSPF code may be in a single area
or in multiple areas.
Model equation gives single Area equation (most used)
A more complex mode adds:
Inter-Area SPF computation
Intra-Area Database Lookup
Intra-Area SPF calculations for AS-External (type-5) and
NSSA Computations
Summary Computations for an Area Border Router
[Summary network (type-3) and Summary ASBR (type-4)]
Benchmarking may give this
With internal sensors, or
For simple situations in black box
10
BGP Convergence
BGP is Path Vector
rumor protocols with paths
BGP is Policy Protocol
Amount policy matters in
convergence
BGP distributes routes from
IGPs and Labels from MPLS
Redistribution times matter
BGP is widely deployed
Data of BGP routes
patterns matter
Processing patterns matter:
Fractal or Markov
11
Multi-AS Convergence for Beacon with
multiple routes
Beacon 1st change 1st change
AS 1 Router1
AS 3
AS 2 Last
Last change
change
Convergence Terms as seen by AS 2 Router 1
•T0 – 1st change received from AS 1
•T1 – 1st change sent to AS 3
•T2 - Last change received from AS 1
•T3 – Last change received from AS 1
Beacon Experiments are collecting data !!
12
Received E-BGP updates
E-BGPec(p) = Ur + (Tr + QSP + RS-Policy + PV+ RIB)(rts) +
IBGPec+ EBGP-lpec + EBGP-ASpec + FIB + CRR
Ur: Update is received
Tr – transmission of Update
QSP: cumulative queuing, serialization,
propagation
RS-Policy: Route Selection Policy
PV: Path Vector Calculation
RIB: RIB is updated
DS-Policy: Distribution Policy
IBGP-ec – IBGP convergence based on exterior
routes
EBGP-lpec – Convergence EBGP Peers for BGP
Update peer local to receiving BGP router
EBGP-ASpec – Convergence of EBGP Peers for
BGP peers within AS
FIB: FIB is updated
CRR – BGP Route Recursion is check
13
E-BGP with no IBGP peers
E-BGPec(p) = Ur(T2-To) + (Tr + QSP + RS-Policy + PV+ RIB)(rts)
+ nEBGP-lpec (DS-Policy + RIB + Uo + Tx) +
nEBGP-Asp-ec(Uo + Tr + QSP + RS-Policy) + FIB + CRR
Ur: A sequence of New Updates are received
Uo – update originated
Tr – processing reception of the New
Updates with Route(s)
QSP: cumulative queuing, serialization,
AS 1 AS 2 AS 3 propagation
RS-Policy: Route Selection Policy
PV: Path Vector Calculation
RIB: RIB is updated
DS-Policy: Distribution Policy
TX – Update is transmitted
FIB: FIB is updated
CRR – BGP Route Recursion is check
14
I-BGP convergence: transit
BGPlec(p) = Ur + N-Peers (Tr + QSP + RS-Policy + PV +
RIB + DS-Policy+ RIB + Tx) + FIB + CRR
D: Failure is detected
LoC(p): IGP convergence
Uo: New Update is originated
Uor: New Update is originated based on E-BGP received •Worse Case =
Tr – transmission of Update Sequential N-
Tx – Update is transmitted
QSP: cumulative queuing, serialization, propagation
Peers
RS-Policy: Route Selection Policy
• Best case = 1
DS-Policy: Route Distribution Policy
PV: Path Vector Calculation
½ peers
FIB: FIB is updated
RIB: RIB is updated
CRR – Route Recursion is check
15
I-BGP convergence: Originated
BGPoc(p) = D + LoC(p) + Uo + N-Peers (Tr + QSP + RS-
Policy + PV + RIB + DS-Policy+ RIB + Tx) + FIB + CRR
D: Failure is detected
LoC(p): IGP convergence
Uo: New Update is originated •Worse Case =
Uor: New Update is originated based on E-BGP received
Tr – transmission of Update
Sequential N-
Tx – Update is transmitted Peers
QSP: cumulative queuing, serialization, propagation
RS-Policy: Route Selection Policy • Best case = 1
DS-Policy: Route Distribution Policy
PV: Path Vector Calculation
½ peers
FIB: FIB is updated
RIB: RIB is udpated
BGP – Route Recursion is check
16
LDP Convergence- IGP/LDP node
LDPoc(p) = D + LoC(p) + LO + (h * (LDP-node))*Path + LIB + LFIB
LDP-node = Lr + LSP-RS-Policy + LD + LSP-DS-Policy + Lx + LIB +
LFIB (preliminary)
D: Failure is detected •Worse Case = All
LoC(p): IGP convergence [From OSPF] paths sequential
LO: New Update to Labels are originated
h – hops for LDP distribution from ingress to egress • Best case = 1 or
Path – Pathways factor 1½ paths
LDP-node – per node distribution
Lr – Label update received
Lx – Label update sent
LIB – label information base updated
LFIB – label forwarding base updated
17
RSVP-TE Convergence times
LoC(p) = RR + RRO + QSP + (h * (rspv-node)) + LIB + LFIB
RR: Reservation Requested
RRO: New RSVP Request
Insertion of External routes occurs here
QSP: cumulative queuing, serialization, propagation,
H*F: LSP is flooded up to rerouting node
SPF: Shortest path calculation
RIB: RIB is updated
FIB: FIB is updated
DD: LCs are updated
CRR: BGP recursion is fixed
18
Benchmarking Performance
Benchmarking: Test topologies
tcpdump tcpdump
DUT TR 2
TR1
•IGP
• Data stream: Generated OSPF LSA/ISIS LSP Sent to node
• Internal measures: SPF convergence, inter-area,
20
Benchmarking: Test topologies
tcpdump tcpdump
DUT TR 2
TR1
•BGP
•Data stream parameters: Prefix mixtures, # of AS in
Path
No substitute for the Real data mixtures
Models: Fractal (blast) or Markov arrival rates
•Policy parameters: RS-Policy and DS-Policy in term of
ACLs or route maps
•TCP Parameters
•BGP protocol: off WRD, IBGP synchronization,
Smoothing, Min-Route Advertisement time
•Measurements: Blackbox (tcpdump) versus Internal
21
Benchmarking: Test topologies
tcpdump tcpdump
DUT TR 2
TR1a TR1b
•L3 VPN
•Data stream: BGP test patterns with Label insertion based
on translation
•Measurement: blackbox (tcpdump) vs internal
•LDP
• Data stream:
•Generated OSPF LSA/ISIS LSP inserted into Node
•LDP generates label per prefix
• Internal measures: SPF convergence, inter-area,
22
Benchmarking: Test topologies
tcpdump tcpdump
DUT TR 2
TR1
•RSVP
•Data stream: Generated Requests based on Traffic
manager model
•External measurement: TCP dump or TR2
23
Q&A
The Internet Continues to Grow
Number of IP v4 Prefixes
In Route table
25
IP v6 is emerging
Number of IP v6 Prefixes
In Route table
26
The Number of Unique Networks (AS) is
growing
27
Networks are getting more connected
Number of AS through
Public Internet is shorter.
28
Less network prefixes per AS
29
