View
4
Download
0
Category
Preview:
Citation preview
SDN with Link-State Routing Protocols
PIRL
Olivier Tilmans
8th March 2016
UCLouvain
Joint work with
S. Vissicchio (UCLouvain), L. Vanbever (ETH Zurich) and J. Rexford (Princeton)
IPv6 comes with a unique opportunity
current network
new network
IPv6 transition
IPv6 comes with a unique opportunity
to improve configuration automation
current network
new network
IPv6 transition
IPv6 comes with a unique opportunity
to simplify the protocol stack
current network
new network
IGPMPLS
RSVPBGP
IPv6 transition
IPv6 comes with a unique opportunity
to change network designs
SDN (e.g., OpenFlow, Segment Routing)
Traditional (e.g., IGP, distributed MPLS)
The state of the art includes two networking models
based on opposite principles
SDN simplifies control-plane and management,
but sacrifices robustness of distributed protocols
Manageability
Flexibility
Scalability
Robustness
SDN
ad hoc
low
highest
high
Traditional
IGP, tunnelling (RSVP-TE)
by design
high
low
low
SDN simplifies control-plane and management,
but sacrifices robustness of distributed protocols
Manageability
Flexibility
Scalability
Robustness
SDN
ad hoc
low
highest
high
Traditional
IGP, tunnelling (RSVP-TE)
by design
high
low
low
We propose Fibbing, a network architecture which
combines advantages of SDN and traditional networking
Fibbingcentral control over a single link-state IGP
SDN with Link-State Routing Protocols
Manageability1
Scalability & Robustness
2 Flexibility
3
SDN with Link-State Routing Protocols
Manageability1
Scalability & Robustness
2 Flexibility
3
achieving central control
A B
C
destinationsource
Consider this simple network
(implemented with Cisco routers)
X
12
A B
C X
An IGP control-plane computes
shortest paths on a shared weighted topology
control-plane
3
1
110
shortest paths
13
IGP shortest paths are translated into
forwarding paths on the data-plane
data-plane
traffic flow
A B
C
X
A B
C X
control-plane
3
1
110
14
In Fibbing, operators can ask
the controller to modify forwarding paths
requirement (C,A,B,X)
A B
C X3
1
110
15
The Fibbing controller injects information on
fake nodes and links into the IGP control-plane
node V1 link (V1,C) map (V1,C) to (C,A)
A B
C X3
1
110
requirement (C,A,B,X)
16
Informations are flooded
to all IGP routers in the network
A B
C X3
1
110
requirement (C,A,B,X)
17
node V1 link (V1,C) map (V1,C) to (C,A)
Fibbing messages augment
the topology seen by all IGP routers
1
A B
C X3
1
110V1
requirement (C,A,B,X)
18
node V1 link (V1,C) map (V1,C) to (C,A)
Augmented topologies translate
into new control-plane paths
A B
C X3
1
110
requirement (C,A,B,X)
1
V1
19
node V1 link (V1,C) map (V1,C) to (C,A)
Augmented topologies translate
into new data-plane paths
A B
C
X
A B
C X3
1
110
1
V1
requirement (C,A,B,X)
20
node V1 link (V1,C) map (V1,C) to (C,A)
SDN with Link-State Routing Protocols
Manageability1
Scalability & Robustness
2 Flexibility
3
Fibbing can enforce
any set of forwarding DAGs
22
paths for the same destination not creating loops
Fibbing can enforce
any set of forwarding DAGs
23
SDN with Link-State Routing Protocols
Manageability1
Scalability & Robustness
2 Flexibility
3
fine-grained control
24
In the following network,
the blue destination is subject to a DoS attack
A C
B
X
D
5
10
5
5
1051
1
scrubber25
Fibbing can steer away
traffic on a per-destination basis
A C
B
X
D
5
10
5
5
1051
1
fA1
requirement (C,D,X) and (B,C,D,X)
26
SDN with Link-State Routing Protocols
Manageability1
Scalability & Robustness
2 Flexibility
3
per-destination load-balancing
27
Leveraging multiple paths is hard
when links/flows have different capacities/demands
A C
B
X
D
10
10
5
5
1010
Higher link bandwidth
Lower link bandwidth
Higher traffic demand
28
Fibbing has fine-grained control over ECMP routing
Adding new equal-cost path
A C
B
X
D
10
10
5
5
1010fA
15
requirement (A,C,X) and (A,B,X) and (A,D,X)
29
requirements splittingRatios{ (A,B): 2 (A,C): 1}
A C
B
X
D
10
10
5
5
1010
fA
15
Fibbing has fine-grained control over ECMP routing
Introducing uneven load-balancing
30
SDN with Link-State Routing Protocols
Manageability1
Scalability & Robustness
2 Flexibility
3
backup paths provisioning
31
Link failures may induce
congestion or increased delays
A C
B
X
D
5
10
5
10
20 20
32
Fibbing can provision backup paths
C
B
X
D
5
10
5
10
20 20
A
fA
15
requirement (A,D,X) asBackupOf (A,B,X)
33
SDN with Link-State Routing Protocols
Manageability1
Scalability & Robustness
2 Flexibility
3
IGP on steroids
We implemented a Fibbing controller
Works with vanilla OSPF and off-the-shelf routers
Supports all presented use-cases
Induces very little overhead on the routers
No impact on SPF computation
IS-IS requires a protocol extension
Source-code on Github
35
By building upon the underlying IGP,
Fibbing is robust and reactive to failures
1. IGP provides sync primitives
2. replica failures have no impact on forwarding
Fibbing naturally supports replicated controllers
1. IGP is sufficient for some failures [Filsfils07]
2. IGP provides a default for partitions
Fibbing easily deals with network failures
36
Fibbing shows the benefits of
central control over distributed protocols
heavy work is still done by routers
IGPs are in charge of all intra-domain paths
Simplifies controllers and improves robustness
network-wide automated control
Realizes SDN management model
Simplifies network design
37
Olivier Tilmans
olivier.tilmans@uclouvain.be
Tell me lies, tell me sweet little lies
— Fleetwood Mac
SDN with Link-State Routing Protocols
fibbing.net
Backup slides
SDN with Link-State Routing Protocols
MPLS+RSVP-TE/SR
can solve all the presented use-cases
Need to provision one tunnel per ingress point
Hard to add/remove equal-paths for elephant flows
Fibbing also controls path cost seen by other protocols
40
Fibbing manipulates IGP topology, so
does MTR, what’s the difference?
MTR is CLI-driven (configuration changes to do on
every router vs flooding)
Cannot do uneven load-balancing
41
Why shouldn’t I use
Policy-Based Routing?
PBR is CLI-driven (configuration changes to do on
every router vs flooding)
PBR decisions are local to a single router
CPU fallback
42
How am I supposed to troubleshoot a
network with fake elements?
The controller is the primary source of information
Fake elements can be quickly identified in LSDBs
43
Openflow solves everything
The controller has to setup flow entries on every switch
Switches rely on the controller to handle failures
IGPs are getting extensions to support Flowspec, …
44
Experiments on real routers show that
Fibbing has very limited impact on routers
1 000
5 000
10 000
router memory (MB)
0.7
76.0
153
50 000
100 000
6.8
14.5
# fake nodes
CPU utilization always under 4%
>> # real routers
45
The controller can choose between a (very) fast
algorithm or one that minimize the augmented topology
Rocketfuel topology of AS1239 (300+ routers)
46
We study which messages to inject
for controlling intra-domain routing protocols
forwarding
paths
weighted
topology
shortest-path
computation
link-state IGP
input function output
47
The output of the controlled protocol
is specified by operators’ requirements
forwarding
paths
weighted
topology
shortest-path
computation
input function
provided by operators or controller optimizers
(e.g., DEFO)
link-state IGP
output
48
Inverse
To control IGP output, the Fibbing controller
inverts the shortest-path function
forwarding
paths
weighted
topology
shortest-path
computation
49
SDN achieves high manageability
by relying on a centralized controller
derives FIB entries
install FIB entries
computes paths
high-level requirements
network controller
configuration
50
Fibbing is as manageable as SDN,
but centralizes only high-level decisions
Fibbing controller
computes paths
requirements
51
Fibbing keeps installation distributed,
relying on distributed protocols
distributed control-plane install FIB entries
computes FIB entries
data-plane52
Distributed installation is controlled
by injecting carefully-computed information
control-plane messages
53
Manageability
Flexibility
Scalability
Robustness
SDN
ad hoc
low
highest
high
Traditional
IGP, tunnelling (RSVP-TE)
by design
high
low
low
Fibbing
by design
high
high
high
Fibbing combines advantages
of SDN and traditional networking
Manageability
Flexibility
Scalability
Robustness
Fibbing
by design
high
high
high
centralized controller
per-destination full control
some functions are distributed
Fibbing combines advantages
of SDN and traditional networking
Our prototype includes algorithms
to compute augmented topologies of limited size
network topology
+
path reqs.
per-destination forwarding DAGs
augmented topology
reduced topology
running network
Compilation Augmentation OptimizationInjection/Monitoring
56
The controller listens through an OSPF adjacency
to keep an up-to-date view of the topology
network topology
+
path reqs.
per-destination forwarding DAGs
augmented topology
reduced topology
running network
Compilation Augmentation OptimizationInjection/Monitoring
57
Recommended