1. it requires an active queue management technique to keep track of the average queue size for a given queue. The average queue size is the metric which monitoring the congestion level in a given queue.
   Instead of dropping packets (like with RED or WRED), ECN markes packets which experience congestion - before its occurance - to signal this information to the source.
2. ECN: to signal persistent congestion
3. no changes in the classical congestion avoidance and congestion control TCP algorithms (slow start, fast retransmit, fast recovery)
4. ECN field: 2 ECN bits (complementary of the 6-bit DSCP)
   1. ECN-capable transport end-systems (ECT codepoints):
      01 (ECT 1) and 10 (ECT 0)
   2. CE codepoint: 11, set if packet would have been otherwise dropped as an effect of active queuing
5. incremental deployment possible
6. signalling:
   1. ECN echo flag (ECE), TCP flag in reserved flag area (bit 9)
      receiver -> source, to signal the occurence of congestion (CE set)
   2. Congestion Window Reaction (CWR): TCP flagin reserved flag area (bit 8)
      source -> receiver, to signal that the congestion notification has been coreclty received and that reaction to congestion started.

Example of router configuration with ECN enabled

Experimental results:

Note: all tests carried out in the local testbed are such that if ECN is disabled RED is still configured. On the contrary, for all the tests in the WAN, if ECN is disabled, tail dropping is used instead of RED, this to clearly understand the basic performance of TCP without either ECN or RED.

Local area test network

• Test A, Baseline ECN testing and configuration
• Test B, Comparison of performance in terms of throughput with and without ECN (TCP streaming mode)
• Test C, Comparison of performance in terms of number of completed session over time for short-lived TCP sessions (request/response mode) with and without ECN
• Test D, Comparison of two individual streams, with without congestion, with non-ECN capable background traffic
• Test E, Comparison of ECN performance for a variable number of TCP streams
• Test F, Analysis of TCP burstiness through snmp as a function of the number of streams and of the ECN configuration
• Test G, Comparison of TCP performance with and without ECN with 2 congestion points

  Wide are test network - Italy/The Netherlands - (RTT = 20 msec)

• Test A, Analysis of ECN performance for a single ECN-capable stream with larger RTT
• Test B, Comparison of TCP performance with and without ECN in case of two congestion points
• Test C, Analysis of TCP performance with ECN of web-like TCP streams