Comparison of the performance for ECN and non ECN capable TCP parallel streams with TWO points of congestion

Goal: to estimate the gain introduced by ECN, in terms of link utilization, in case of TCP streams (with and without ECN) that across two points of congestion. And also to estimate the actual burstiness degree produced by this TCP concurrent streams by monitoring the instantaneous/average queue length .

Equipment:

• chain of intermediate ecn-capable routers (Cisco 7500 and C7200), IOS: 12.1(4)E (experimental version) – first and second  router are ecn-capable in this test with the same configuration
• pool of Linux hosts, v. 2.4.9, SACK-capable
• poll of Solaris hosts, v. 2.7, non SACK-capable

Test description:

1. Topology see Test A.
2. Parameters
• min threshold constant = 100 pack
• max threshold constant = 300 pack
• drop probability: [1/50, 1/5, 1/2]
• TCP stacks: ECN and SACK-capable and non-ECN and SACK-capable
•  test duration: 3 min 
3. Traffic profile
• 90 TCP streams generated by iperf:
• 30 TCP streams between two Linux hosts, ECN and SACK capable;
• 30 TCP streams between Solaris (Sunlab3) and Linux hosts, non ECN and non SACK capable
• 30 TCP streams between Solaris (Sunlab2) and Linux hosts, non ECN and non SACK capable

4.  Test methodology
   SNMP has been used to query the ECN capable router every 2 seconds. Two are the variables tracked: SnmpCurrQ, SnmpMeanQ, both from the Cisco-class-based-qos-mib.
   When monitoring the two parameters we verified that only every 8 or 10 seconds the two parameters are updated, this means that 8-10 sec is the minimum monitoring granularity achieved.
   This is the script used to query the router

Summary:

• with two points of congestion, ECN  produces a considerable increase in link bandwidth utilization compared to non ECN capable streams
• we can see that with ECN the overall burstiness decreases if it is compared to that produced by non ECN capable streams

Comments:

·        As shown in Figure 1, with ECN the overall link bandwidth utilization is higher than non ECN capable streams for different mark probability and with two points of congestion. It can be a proof of the benefit of ECN for high congested networks. The maximum gain  is obtained for mark probability equal to 1/5, a medium value.

·        From a comparison of the instantaneous queue length with and without ECN, we can see that with ECN the overall burstiness is still reduced, as in the case of only one point of congestion. This confirm the correct behavior of ECN