Measuring the performance of an active network measurement platform

1
Measuring the performance of an active network
measurement platform
Timo-Pekka Heikkinen TKK Networking
laboratory Supervisor Prof. Raimo
Kantola Instructor Lic. Tech. Marko Luoma
2
Contents

• Theory part
• Terms and concepts
• Practical part
• Brix system
• Results
• Conclusions
• Future work

3
Theory part

• Active measurements in data communication
  networks
• Active vs. passive measurements
• Types of active measurement mechanisms and
  methods
• Difficulties in making active measurements
• Accuracy of measurements

4
Terms and concepts 1

• Probe
• An artificial packet containing information
  relevant to the measurement (e.g. timestamp, seq.
  etc.)
• One-way latency (delay)
• Time it takes for a probe to travel from point A
  to point B in a network
• Round-trip delay
• Time it takes for a probe to travel from A to B
  and back to A
• Delay jitter
• The amount of variation in the measured delay
• Packet loss
• A packet is declared lost when it has failed to
  arrive to its destination in a certain amount of
  time

5
Terms and concepts 2

• Types of delay
• Processing delay
• Time it takes for a router to process a packet
• Transmission delay
• Time it takes for a router to push a packet to
  the link
• Propagation delay
• Time it takes for a signal to travel across the
  physical medium
• Queuing delay
• Time the packet spends inside routers queues

6
Practical part

• Measuring the performance of an active
  measurement system (Brix Networks)
• Two test cases / network environments
• Simple network topology
• Comparison of the Brix system versus an
  well-known accurate measurement system (AX4000)
• (Relatively) Complex network topology
• Is Brix able to detect certain events in the
  network?
• Overall, how does the Brix system perform?

7
Devices

• Brix 1000 Verifier

• Synchronization
• Brix 1000 GPS-receiver
• Acted as an NTP-server
• Brix 100s NTP synched
• NTP synch in the complex case
• No need for GPS because RTT was measured

Brix 100 Verifier
8
Brix Architecture
9
Brix Reporting
Data from the Brix system
10
Results (1/4), case 1
Cumulative probability distributions of all
tested devices when measuring one-way delay.
Delay shown with and without NTP-offset correction
11
Results (2/4), case 1
Measured one-way delay between Brix 1000 and Brix
100 verifiers
12
Results (3/4), case 2
Effects of high load on the delay measured by the
Brix system
13
Results (4/4), case 2
Core router failure as reported by the Brix
system.
14
Conclusions (1/2), case 1

• More time should have been spent on designing the
  test setup
• Brix 100s are not accurate enough to measure
  sub-millisecond one-way delays
• Clock instability big issue
• Accuracy 940 µs, 120 µs with NTP corr.
• Brix 1000s are accurate when using GPS
  synchronization
• Clocks stabile enough
• Accuracy 50 µs, 10 µs with NTP corr.
• AX4000 accuracy 1 microsecond (when using only
  one clock)

15
Conclusions (2/2), case 2

• Test parameters should have been selected more
  carefully
• Measurement reports are lost too often
• Congestion and node failures can be detected with
  Brix
• Short link failures were undetected

16
Future work

• Case 1 should be done in a more complex network
  environment
• Case 2 should be done with the same equipment as
  used in case 1
• GPS synchronized Brix 1000s
• NTP offset measurement