Transcript slides - Network and Systems Laboratory

On the Interaction between
Dynamic Routing in the
Native and Overlay Layers
Infocom2006
Srinivasan Seetharaman and Mostafa Ammar
College of Computing Georgia Institute of
Technology
Presenter: Elaine
1
Roadmap
• Introduction
• The model adopted for analysis
• The characteristics of Dual Rerouting
• Layer aware rerouting algorithms
• Simulation result
• Tuning the native layer routing protocol and novel approach
• Conclusion
2
Introduction(1/3)
• Overlay network
• Interaction between two routing layers
– Dynamic routings both on overlay and native IP
network cause problems
– Focus on the specific problem of rerouting
around failed native link
3
Introduction (2/3)
• Mix dynamic routing environment can be avoided?
– Not using dynamic routing in the overlay network and
always counts on the native networks to re-configure
the route?
Overlay dynamic routing can
significantly enhance the overlay
network’s survivability
4
Introduction (3/3)
• Three contributions
– Provide the understanding of Dual Rerouting
– Three approaches to mitigate the rerouting
problems
– Motivate the need for an overlay-aware native
network
5
Rerouting Model(1/5)
A. Framework
– Two scenarios
• Single-Domain Overlay over Single Domain
native
• Single-Domain Overlay over Multiple-Domain
Native
6
Rerouting Model(2/5)
– Generic parameters of each routing protocols
• Cost
• Assume the two ends of each link (native & overlay)
use a keepAlive protocol for link verification.
– 3 keepAlive messages lost  Failure
• A keepAlive time: Time between two keepAlive
messages
• A hold time: Time window to declare link as down
7
Rerouting Model(3/5)
B. Rerouting schemes
– No awareness
• Dual Rerouting
• The benchmark for performance comparison
– Awareness of lower layer’s existence
– Awareness of lower layer’s parameters
8
Rerouting Model(4/5)
C. Performance metrics
1.
Hit-time: Time taken for traffic to be recovered.
2.
Success rate of recovery
3.
Number of route flaps
4.
Peak & Stabilized inflation (before repair)
= Detection time
+
(depends on timers)
Success rate of a layer =
Average route flaps =
Path cost inflation =
Convergence time
(protocol specific)
+
Device time
(Negligible)
Number of paths recovered
Number of failed overlay paths
Number of route flaps
Number of failed overlay paths
Path cost after rerouting
Path cost before failure
9
Rerouting Model(5/5)
10
Characteristics of Dual
Rerouting
• Each layer operates completely independent of the other
– lead to a large number of route flaps and increase the path
cost inflation
• To mitigate this problem
– Vary Hold-time & KeepAlive-time at overlay layer to determine
at which layer detection is likely to happen first
• Best Hit time
– According to the definition of Detection
• Using layer-aware scheme to trade off improvement in
other metrics with longer hit-times.
11
Layer Aware Overlay
Rerouting(1/2)
• These schemes require knowledge of the native layer’s routing
protocol existence or some minimal knowledge
• Probabilistically Suppressed Overlay Rerouting
– suppression operation is done with probability p on each overlay
rerouting attempt
– P=0, dual rerouting
– P=1, native rerouting
– Decrease the number of flaps and achieve better path cost
– Longer hit time
• Deferred Overlay Rerouting
– Delay overlay recovery by a constant value
– After that time has elapsed, if the native network has not yet recovered,
overlay recovery is performed
– fewer route flaps relative to Dual Rerouting
– Longer hit time
12
Layer Aware Overlay
Rerouting(2/2)
• Follow-on Suppressed Overlay Rerouting
– The overlay layer keeps track of the native
layer’s timer values
– Follow-on time
• If follow-on time < threshold , then suppress overlay
rerouting
• Similar to DOR, but has a relatively smaller hit-time
13
Result
A. Simulation Setup
–
–
–
–
–
5 overlay topologies over 5 native topologies =>25
combinations
Native layer
•
Intra-domain
•
Inter-domain
–
keepAlive time(1s)/ KeeAlive message(3)/ Hold-time(3s)
–
keepAlive time(5s)/ KeeAlive message(3)/ Hold-time(15s)
Overlay layer
•
2 or 3 keepAlive message
•
•
Intra-domain(10 overlay nodes/100 native nodes)
Inter-domain(10 overlay nodes/500 native nodes)
•
stateless all-link failure approach
Network topology
Link Failure Modeling
14
Dual Rerouting
• Ovelay hold time↑ Hit time ↑
15
• Ovelay hold time↑ Avg Route flaps ↓
• Show trade off between Hit time & avg route
flaps↓
16
• Overlay hold time↑ Peak inflation ↓
• Peak inflation, native-only is the best
• Stable inflation, irrespective to hold time
17
Performance Comparison
– Hold time : 3.0s
– The delay and follow-on threshold were set to 0.375 (2 secs)
– the suppression probability was set to 0.5
1. Native-only rerouting must attain steady state faster than the other
2. Dual Rerouting, which has no suppressed overlay rerouting
18
operations, must attain the peak earliest
19
Summary of performance
of layer-aware schemes
• None of the three layer-aware schemes
are the best
– Based on what the system is sensitive to ,
different scheme can be chosen
• In most situations, packet loss has a more
serious effect on the performance of the
overlay traffic
– Reducing the hit-time should be given a high
precedence
20
TUNING NATIVE LAYER
PARAMETERS
• Motivation
– overlay applications proliferate, the native layer should
gradually evolve to suit the overlay network requirements
– Improve the performance by adjust parameters of the native
layer routing protocol
• Tuning the keepAlive-time
– Native layer rerouting was shown to be the optimal one in
terms of path cost inflation and number of route flaps
– In Dual Rerouting, insuring that recovery will take place at the
native layer first can be achieved by setting the native layer’s
keepAlive-time to a value much smaller than that at the
overlay layer
21
• We define the routing protocol overhead as the
number of keepAlive packets sent per second on
the link under consideration
22
23
• Summary of performance gain with native
layer tuning
– Help Dual Rerouting improve the performance
– Maintains the overall routing overhead the same
– Reducing the keepAlive-time at the native layer also
benefits the non-overlay applications sharing it
24
Conclusion
– Provide the understanding of Dual
Rerouting
– Three approaches to mitigate the
rerouting problems
– Motivate the need of tuning the keepAlivetime of the native layer to achieve the best
possible rerouting performance
25
26