Preliminary simulation results on differentiated QCN

1
Preliminary simulation results on differentiated
QCN

• Marina Lipshteyn

2
The scenario

• 4 Reaction points which transmitted constantly
  3.8Gbs to 1 congestion point with pkts size -
  0.5KB.
• Among them, one victim TCP session.

Victim flow TX 3.8Gbs
CP
3.8Gbs
3.8Gbs
3.8Gbs
3
With PFC, without QCN

• The incoming traffic is reduced to

2.3Gbs
2.3Gbs
2.3Gbs
2.7Gbs
4
With PFC and QCN
The incoming traffic is reduced to
3.8Gbs
1.6Gbs
2.3-1.8Gbs
2.3-1.8Gbs
5
Using different types of traffic

• Only one of the Reaction Points is injecting
  traffic above commited rate.
• In my opinion, it is more correct to pause that
  Reaction Point only.

policer
Victim flow TX 3.8Gbs
policer
policer
policer
Overcommited traffic
CP
3.8Gbs
3.8Gbs
3.8Gbs
6
Differentiated QCN at the Congestion Point

• At the Congestion Point define two thresholds for
  the queue
• The first threshold
• When number of bytes in the queue exceeds it,
• send CNM ONLY IF the probed packet is YELLOW.
• The second threshold
• When number of bytes in the queue exceeds it,
• send CNM for any probed probed packet

7
With PFC and differentiated QCN
The incoming traffic is reduced to
3.8Gbs
2.3Gbs
1.6Gbs
3.8Gbs
Overcommited traffic