Demonstrate Cascading Failures in ns
Xuan Chen, Ashish Goel, John Heidemann, Ramesh Govindan, and Deborah
Estrin
Mar 28, 2002
Goal:
- Understand, detect, predict, and avoid network cascading failures;
- Cascading failure is a disruption at one point in the networks
causes a disruption other points;
- Cascading failure could be caused by the sudden shift in network
traffic and/or the propagation of router failures.
Current stage:
- Demonstrate network cascading failure scenarios under simulations in
ns (one example is given below)
- Investigate metrics to quantify failures (such as the RTTs of cross
traffic)
Future plan:
- Verify simulation results with network traces
- Understand network cascading failure scenarios from network traffic
perspective
- Implement mechanism to detect and predict network cascading failures
Sample Scenario:
Traffic in the simulation:
- Pulse UDP: ON/OFF traffic with short
ON period but high packet rate within the ON period;
- Low rate CBR: UDP traffic with a small
but constant packet rate;
- TCP cross traffic: TCP
flows share no common links with the pulse UDP flow.
Simulation snapshots:
Initial state
- Low rate CBR does not affect
TCP cross traffic flows
- No large queue or packet loss is observed
|
|
Pulse occurs
- Pulse UDP flow starts
- Packets in low rate CBR are blocked
|
|
Pulse "spreads"
- Pulse UDP flow ends
- Packets in the low rate CBR flow
, which are queued, form a secondary pulse
|
|
Pulse hurts cross traffic
- The secondary pulse hurts cross
traffic
- Large queue and packet loss are observed
|
|
Last update: Thu
Mar 28 11:15:07 PST 2002
Mail any comments, suggestions to
Xuan Chen