Our goals of the first experiment:

• Verify that relevant network service performance parameters (delay, jitter, loss) on a segment and end-to-end can be obtained using packet capturing method
• Verify that faults can be detected and localized on a network segment
• Estimate scalability of the proposed solution or the need to balance between result accuracy and scalability by sampling captured traffic. This will be done by estimating the growth of the traffic towards the domain collector, storage space, memory and CPU consumption of the domain collectors with the growth of the user traffic, number of service instances and packet capturing points.  
• Estimate the impact of VPN topology changes on packet loss detection and define the algorithm for detecting topology changes

Software that we have to install/mechanisms to define (please suggest)

• Accurate time at packet capturing nodes. What is the software which will give us adequate time accuracy and time resolution? Although VMs are probably not the best machines to keep accurate time, we should try to install the software which will be used in our final solution.
• Traffic generation software:
  • tcpreplay
• Traffic capturing software (capable to capture all the significant parts of a packet): just tcpdump/tshark or whatever at the very begining (in order to emphasize their limits)?
• Mechanism to transport captured traffic from packet capturing points to the management station: just rsync or whatever at the very begining?
• Software needed to emulate delay, jitter, loss
  • netem – check if this is sufficient for us, and if it can be installed/used
• Compare our statistics with basic active measurements (that can be also sniffed)?

We have to make:

• Addidional node configuration to match all our experimentation goals (additional VLANs, define final testbed and service topology)
• Application/script which extracts only the necessary data from the packet capture
• Application/script which transports captured data to the central management station
• Define method and create and application/script which analyses gathered results and provides results in the human readable form (no nice GUI though).

We can assume:

• That at the moment the topology of all service instances is known by the domain controler. We will later try to solve the problem of extracting service instance topology from the network, and adapting performance calculation to network topology changes

We will work later:

• On the specification of the application, data models and so on...

First experiment

The results of the first experiment are shown here.

JRA2T4 Lab Testbed

Details about the testbed can be found here