Terminology for Benchmarking Network-layer Traffic Control Mechanisms - draft 4

1
Terminology for Benchmarking Network-layer
Traffic Control Mechanisms - draft 4

• Jerry Perser
• David Newman
• Sumit Khurana
• Shobha Erramilli
• Scott Poretsky

55th IETF Atlanta, USA
2
Changes

• Added new references
• Jitter 6 and 7
• Congestion 8
• Edited Jitter Definition
• Jitter is between packets in the same stream.
• Clarified differences between IPDV (RFC 3393)
• 54th IETF meeting notes
• The meeting agreed that it was ready for WG last
  call.

3
Congestion

• A condition in which one or more egress
  interfaces are offered more packets than are
  forwarded.
• Loss can determine if offered gt forwarded
• Delay may or may not be caused by congestion
• Offered packets to egress interface excludes
  traffic destined to the DUT.
• The DUT does not forward packets going to itself.

4
A Congestion Test

• DUT is Fast Ethernet layer 2 switch
• Throughput is line rate.
• Two ports of 64 byte packets to a single port
• Single class of service (one queue)
• OLOAD set from 0 to 100 in 0.1 increments
• Test duration 60 seconds
• Predicted results?

DUT
5
Average Delay from 0 to 8
1 Packet delay 5.76uS
1.8uS
1.1uS
6
Average Delay from 0 to 46
Point of Inflection
7
Average Delay from 34 to 50
Another Point of Inflection
8
Average Delay from 0 to 100
Last Point of Inflection
No Loss lt 50
Loss gt 50
9
Three Points of Inflection

• Ignoring the singles delay increases
• Sloped of the Average delay changes
• 46.5 no loss
• 49.6 no loss
• 50.1 LOSS
• Only Loss matched the predicted congestion point

10
To Do List

• Change reference number 6
• draft-ietf-ippm-ipdv-10 is now RFC 3393
• Add paragraph to Congestion Discussion
• Why delay is not deterministic indicator of
  congestion
• Comments on the difference between delay and RFC
  2679
• RFC 2679 measurement varies with packet size

11
Any Questions ?
The Gun Fighter 1917