Network Simulation and Testing

1
Network Simulation and Testing

• Polly Huang
• EE NTU
• http//cc.ee.ntu.edu.tw/phuang
• phuang_at_cc.ee.ntu.edu.tw

2
Testing on Real Systems

• dummynet

3
4 Weeks

• Week 1
• Introduction, basic features
• Week 2
• Setup and bandwidth management
• Week 3
• Advanced features
• Week 4
• Testing over a simple network

4
Today

• Advanced features
• Largely based on Luigis slides

5
Quick Review of Basic Features

• Two major components
• The pipe
• The packet filter
• Simple examples
• Configuration
• System setting

6
A Simple Example
out
IP layer
1
2
in
Link layer
ipfw add pipe 1 ip from any to any in ipfw add
pipe 2 ip from any to any out
7
The Detailed System View
Higher-level processes
in-pipe
ipfw
else drop
true
true
out-pipe
ipfw
else drop
in-traffic
out-traffic
8
Advanced Features

• Multipath
• For packet re-ordering emulation
• Dynamic pipes
• For automatic per-flow piping
• Dynamic queues
• For automatic per-flow queuing
• Extensions
• Bursty loss and delay jitter

9
Multipath

• For emulation of packet re-ordering
• The usage
• Create multiple paths
• Filter packets to these paths based on certain
  criteria
• Currently, the criteria is probabilistic

10
Multipath Example
ipfw pipe 1 config bw 2Mbits/s delay 80ms ipfw
pipe 2 config bw 3Mbits/s delay 30ms ipfw pipe 3
config bw 1Mbits/s delay 100ms ipfw add prob 0.33
pipe 1 tcp from me to any ipfw add prob 0.5 pipe
2 tcp from me to any ipfw add prob 1 pipe 3 tcp
from me to any
11
The Rules of Multipath Filtering

• For each packet arriving at the dummynet
• Match the 1st filter
• For each packet arriving at the filter
• Take a random number
• Check if it falls within the filter probability
• If yes, pass packets through the rule pipe
• If not, pass packets to the next filter
• Recursive use of the packet arriving at the
  filter algorithm

12
Illustration of the Multi-Filtering
Higher-level processes
0.67
Filter 1
0.33
Filter 2
0.5
0.5
Filter 2
0
1
else drop
Pipe 1
Pipe 2
Pipe 3
out-traffic
13
Dynamic Pipes

• For per-flow piping and monitoring
• The usage
• Adding a mask to the source or destination IP
• Representing the level of flow classification

14
Dynamic Pipe Example
ipfw pipe 1 config bw 2Mbits/s delay 80ms mask
src-ip 0x000000ff
15
The Rules of Per-Flow Piping

• mask src-ip 0x000000ff
• Packet src/dst addresses are ANDed with the mask
• Src/dst addresses masking to the same number
  belong to the same flow
• 256 flows in this example
• The number is the flow ID
• A pipe of the same configuration is created per
  flow

16
WFQ

• Weighted Fair Queuing
• For sharing of bandwidth
• Statistical limitation
• One pipe for all flows
• A flow-based queue to determine which flow gets
  how much bandwidth share

17
WFQ Example
ipfw pipe 10 config bw 2Mbits/s ipfw queue 1
config pipe 10 weight 50 ipfw queue 2 config pipe
10 weight 20 ipfw queue 3 config pipe 10 weight
1 ipfw add queue 1 tcp from me to any ssh ipfw
add queue 1 udp from me to any 53 ipfw add queue
2 tcp from me to any 25 ipfw add queue 3 IP from
me to any
18
Calculating the Share

• Sum all the weights
• The flow gets the fraction of bandwidth
  proportional to
• The ratio of its weight to the sum

19
Statistical Sharing

• Note the distinction of statistical bandwidth
  limitation to deterministic
• One flow can get the whole bandwidth if no one
  else is in queue

20
Dynamic Queuing

• The idea is similar to dynamic piping

ipfw queue 1 config pipe 10 weight 50 mask src-ip
000000ff
21
Extension

• What if we want to emulate
• Bursty losses
• Delay jitter

22
Using Existing Mechanism

• Bursty losses
• Add bursty traffic to compete with the traffic of
  observation
• So that the bursty traffic cause bursty drops to
  the traffic of observation
• Delay jitter
• Add varying traffic to cause varying queue size
• So that the delay variance is of the traffic of
  observation goes up

23
Adding to dummynet

• Modify
• ip_fw.c
• ip_dymmynet.c
• The BSD kernel is
• Very modular
• Extensively commented

24
Performance

• Packet forwarding rate with pipe is a few lower
  than without pipe
• Packet copying is constant
• Queue and pipe schedule is O(logN)
• N number of queues or pipes
• Pipe/queue state fetching could be time consuming

25
Conclusion

• Quite easy to use
• If installing BSD isnt a problem
• If loading the dummynet kernel isnt a problem
• Quite a bit of applications with the simple
  functions
• Need to be careful about how data flow
• Complicated applications might be achievable with
  a bit of creativity
• Usually the hardest part is knowing what to test

26
Next Week

• Laptops
• Running FreeBSD
• With access to FreeBSD
• FreeBSD
• The kernel with proper options