Governing Routing in the Evolving Internetcompunet/www/docs/Vissicchio... · contribution overview...
Transcript of Governing Routing in the Evolving Internetcompunet/www/docs/Vissicchio... · contribution overview...
Governing Routing
in the Evolving Internet
Candidate: Stefano Vissicchio
Advisor: prof. Maurizio Pizzonia
PhD dissertation 19/04/2012
Governing Routing
in the Evolving Internet
network management is necessary
clean-slate design is rare
… crucial …
large businesses lose 3.6% (on avg.) of annual revenue due to network downtime [infonetics04]
downtime costs several millions of USD/h for critical apps [YankeeGroup04]
… and difficult
heterogeneous devices running interacting distributedprotocols
configuration languages are low-level, few automation
50-80% of network outages are due to human errors [juniper08]
Governing Routing
in the Evolving Internet
we take the perspective of a single ISP
Border Gateway Protocol (BGP)
Internet Service
Provider (ISP)
Governing Routing
in the Evolving Internet
routing traffic flows $$$
ISP XeBGPsession
eBGPsessions
eBGPsession
IGP + iBGP
Governing Routing
in the Evolving Internet
network evolution is needed
update the network infrastructure
new technologies, topology changes, etc.
protocol dynamicity
e.g., routing table growth
satisfy new requirements
better performance, new services, etc.
achieve competitive advantages
adapt to traffic requests, reduce costs, etc.
Governing Routing
in the Evolving Internet
iterative methodologies [Oppenheimer04,Teare07]
PLAN
DESIGN
DEPLOYOPERATE
OPTIMIZE
time
*
change
needed
Summary of the
Thesis Contributions
span different phases
network monitoring
assess the state of the
network
PLAN
DESIGN
DEPLOYOPERATE
OPTIMIZE
Summary of the
Thesis Contributions
span different phases
network monitoring
assess the state of the
network
configuration testing
static testing for runtime
properties
PLAN
DESIGN
DEPLOYOPERATE
OPTIMIZE
Summary of the
Thesis Contributions
span different phases
network monitoring
assess the state of the
network
configuration testing
static testing for runtime
properties
reconfigurations
lossless configuration
changes
PLAN
DESIGN
DEPLOYOPERATE
OPTIMIZE
Use Case:
Dealing with a Change
time
*change
needed
*change
applied
initial
routing state
final
routing state
Use Case:
Dealing with a Change
time
*change
needed
*change
applied
initial
routing state
final
routing state
Reconstructing
the Initial State
problem: monitor the network
without interfering with
network operation
contribution overview
look into both control- and data-plane
traffic matrices computation and protocol debugging
propose new monitoring solutions
exploit advanced router features
experiment and evaluate
PLAN
DESIGN
DEPLOYOPERATE
OPTIMIZE
Existing Approach Accuracy vs Overhead
direct measure too much overhead [Zhao06]
packet
samplingtradeoff between accuracy and overhead
[Netflow9, Feldmann00, Papagianaki04]
estimation limited accuracy [Medina02, Varghese04, Zhao06]
Computation of
Traffic Matrices [Brock95]
proposal*: leverage router programmability
install optimized packet counters, and keep them
aligned to routing in programmable routers
>99% accuracy for 98% of IP prefixes
low overhead and performance degradation
* Balestra et al. “Leveraging Router Programmability forTraffic Matrix Computation”, PRESTO 2010
Debugging of
Routing Protocols
proposal*: exploit selective packet cloning
packets are copied by routers’ hardware
all data available to a centralized collector
router performance degradation within 2%
* Vissicchio et al. “Beyond the Best: Real-Time Non-Invasive Collection of BGP Messages”, INM 2010.
Existing Approach Collecting Abilities vs Intrusiveness
push techniques[Quagga, Pyrt]
limitations on collected data,
not in real time
pull techiques[SNMP, scripts]
routers performance degradation,
not in real time
ad hoc protocols[BMP]
impact on routers performance,
required device support
Use Case:
Dealing with a Change
time
*change
needed
*change
applied
initial
routing state
final
routing state
Routing
Configuration Design
problem: guarantee correctness
convergence to a stable state [Griffin02]
no forwarding loop [Griffin02]
contribution overview*
formalization of a third correctness property
each router should have a route
discussion of practices to tweak iBGP
more flexibility less correctness guarantees
proposal of design guidelines
PLAN
DESIGN
DEPLOYOPERATE
OPTIMIZE
* Vissicchio et al. “iBGP Deceptions: More Sessions, Fewer Routes”, INFOCOM 2012; Cittadini et al.
“Doing Don'ts: Modifying BGP Attributes within an Autonomous System”, NOMS 2010.
Routing
Configuration Testing
problem: check for routing convergence
guarantees
policy-based protocols (BGP) can
[Griffin99] and do [Berger01] oscillate
assumption-based tools [Flavel08]
PLAN
DESIGN
DEPLOYOPERATE
OPTIMIZE
contribution overview
deep theoretical study
complement state of the art [Griffin02,Feamster05,Sami10]
practical techniques and tools
based on the gained insight
State of the Art:
Problems and Conditions
NO
DISPUTE
WHEEL
NO
DISPUTE
RING
HAS A
STABLE STATE
(NP-hard)
SUF
SAFE
A New Stricter
Sufficient Condition*
HAS A
STABLE STATE
NO
DISPUTE
REEL
(NP-hard)
SAFE
SUF
* Cittadini et al. “Wheel+Ring=Reel: the Impact of Route Filtering on the Stability of Policy Routing”,
ICNP 2009 andTON 2011
… But All Problems are
Computationally Hard*
HAS A
STABLE STATE
SUF
SAFE
NO
DISPUTE
REEL
(NP-hard)
(coNP-hard)
(coNP-hard)
(coNP-hard)
* Chiesa et al. “LocalTransit Policies and the Complexity of BGP StabilityTesting”, INFOCOM 2011
A Heuristic-Based
Practical Approach*
translation to a
generic model
provably correct
(not complete)
heuristic
* Cittadini et al. “FromTheory to Practice: Efficiently Checking BGP Configurations for Guaranteed
Convergence”, TNSM 2011
A Heuristic-Based
Practical Approach*
flexibility scalability
* Cittadini et al. “FromTheory to Practice: Efficiently Checking BGP Configurations for Guaranteed
Convergence”, TNSM 2011
Use Case:
Dealing with a Change
time
*change
needed
*change
applied
initial
routing state
final
routing state
Network-Wide
Routing Reconfigurations
problem: change routing configuration with
no service interruption
to support stringent SLAs
routing and forwarding anomalies can
occur in intermediate states [Vanbever11]
PLAN
DESIGN
DEPLOYOPERATE
OPTIMIZE
contribution overview
model and theoretical studies
first of its kind for both IGP and BGP
provable correct methodologies
prototype provisioning system
Network-Wide
Routing Reconfigurations
Existing Approach Limitations
current best practices
[Pepelnjak07,Herrero10]
only rules of thumb,
no provable methodology or tool
Shadow Configuration
[Alimi08]not applicable today
Graceful Operations
[Keralapura06,Raza11]case-specific, hard to generalize
proposals*: methodologies and prototype tools
study the operational ordering problem
computationally hard but algorithmically solvable in IGP
requires external help (e.g., VRFs) in BGP
evaluation in virtual environments
no packet loss can be achieved
* Vanbever et al. “Seamless Network-Wide IGP Migrations”, SIGCOMM 2011 and TON 2012
Architecture of the
IGP Provisioning System
controls
intermediate
routing statesensures no
packet loss
interacts with
routers
Evaluation of the
IGP Provisioning System
Publications:
Journals
1. Vanbever, Vissicchio, Pelsser, Francois, Bonaventure. Lossless Migrations of Link-State IGPs. IEEE/ACM Transactions on Networking, 2012. to appear.
2. Cittadini, Di Battista, Rimondini, Vissicchio. Wheel + Ring = Reel: the Impact of Route Filtering on the Stability ofPolicy Routing. IEEE/ACM Transactions on Networking. 19(4):1085 – 1096, 2011.
3. Cittadini, Rimondini, Vissicchio, Corea, Di Battista. FromTheory to Practice: Efficiently Checking BGP Configurations for Guaranteed Convergence. IEEE Transactions on Network and Service Management, 2011
Publications:
Conferences
1. Vissicchio, Cittadini, Vanbever, Bonaventure. iBGPDeceptions: More Sessions, Fewer Routes. In IEEE INFOCOM, 2012.
2. Vanbever, Vissicchio, Pelsser, Francois, Bonaventure. Seamless Network-Wide IGP Migrations. In ACM SIGCOMM, 2011.
3. Chiesa, Cittadini, Di Battista, Vissicchio. Local TransitPolicies and the Complexity of BGP Stability Testing. In IEEE INFOCOM, 2011.
Publications:
Conferences
4. Balestra, Luciano, Pizzonia, Vissicchio. Leveraging Router
Programmability for Traffic Matrix Computation. In
ACM PRESTO 2010.
5. Vissicchio, Cittadini, Pizzonia, Vergantini, Mezzapesa,
Papagni . Beyond the Best: Real-Time Non-Invasive
Collection of BGP Messages. In INM/WREN, 2010.
6. Cittadini, Di Battista, Vissicchio. Doing Don'ts: Modifying
BGP Attributes within an Autonomous System. In
IEEE/IFIP NOMS, 2010.
Publications:
Conferences
7. Cittadini, Di Battista, Rimondini, Vissicchio. Wheel +
Ring = Reel: the Impact of Route Filtering on the
Stability of Policy Routing. In ICNP, 2009.
Publications:
Technical Reports
1. Vissicchio, Cittadini, Vanbever, Bonaventure. iBGPDeceptions: More Sessions, Fewer Routes. RT-DIA-189-2011, Roma Tre, 2011.
2. Vanbever, Vissicchio, Pelsser, Francois, Bonaventure. SeamlessNetwork-Wide IGP Migrations. UCLouvain, 2011.
3. Cittadini, Mezzapesa, Papagni, Pizzonia, Vergantini, Vissicchio. Beyond the Best: Real-Time Non-Invasive Collection of BGP Messages. RT-DIA-165-2010, Roma Tre, 2010.
4. Pizzonia, Vissicchio. Test Driven Network Deployment. RT-DIA-143-2009, Roma Tre, 2009.
Thank You !
Questions ?