Information about testbed configuration

• Hardware and software configuration of router gsr-ipv6
• pool of 4 Linux 7.1 (kernel 2.4.17 hosts): end-node configuration
• baseline measurement of RTT between source (tcp1-ge.nordunett.no) and destination (tcp4-ge.nordunett.no):

  
  --- tcp4-ge.uninett.no ping statistics ---
  100 packets transmitted, 100 packets received, 0% packet loss
  round-trip min/avg/max/mdev = 0.166/0.223/0.286/0.034 ms
  

Experimental results

• Test 1: Performance of Eng 2 and Eng 3 interfaces for different levels of traffic load and congestion debugging; Results

• Test 2: Analysis of MDRR traffic isolation in presence of 2 classes; Results

• Test 3: Analysis of MDRR traffic isolation in presence of 5 classes and with and without WRED; Results

• Test 4: DRR and low-latency (stric priority) queuing delay and jitter properties

  Note: in this experimental set-up synchronization of end-system clocks is achieved with NTP; with the result that one-way delay is substantially over-estimated and in particular, the absolute value of one-way delay varies from test to test.
  In what follows we will concentrate on the one-way delay behavior, for example its frequency distribution function rather than on absolute delay values. On the other hand, IPDV absolute values and distributions are comparable from test to test.

  • Test 4.1: baseline delay and IPDV performance; Results

  • Test 4.2: delay and IPDV performance of MDRR with offered load per-class equal to the bandwidth assigned to the that class (no over-estimation, 5 queues): Results

  • Test 4.3: delay and IPDV performance of MDRR when the configured bandwidth of a given reference class is two times the offered load of that traffic class (over-estimation factor: 2); Results

  • Test 4.4: delay and IPDV performance of MDRR when the configured bandwidth of a given reference class is three times the offered load of that traffic class (over-estimation factor: 3); Results

  • Test 4.5: comparison of low latency queuing and MDRR; Results

    Summary table: IPDV distribution

    Test Num	Type of scheduling configuration	IPDV Frequency distribution (4 sec)
    -------------------------------------------------------------------------------------------
    4.1		No scheduling, no congestion		5%, 85%, 10%
    4.2		MDRR, no over-estimation		2.5%, 45%, 30%, 20%, 2.5%
    4.3		MDRR, over-estimation factor: 2		1%, 72%, 26%, 1%
    4.4		MDRR, over-estimation factor: 3		6%, 72%, 22%
    4.5		LLQ					36%, 61%, 3% 
    
    		

    Summary table: max IPDV

    Test Num	Type of scheduling configuration	Max IPDV (msec)
    							1 sec		4 sec
    -----------------------------------------------------------------------------	
    4.1		No scheduling, no congestion		0.124		0.138
    4.2		MDRR, no over-estimation		0.104		0.105
    4.3		MDRR, over-estimation factor: 2		0.111		0.131
    4.4		MDRR, over-estimation factor: 3		0.107		0.138
    4.5		LLQ					0.106		0.169
    
    

• Test 5: analysis of buffer allocation in buffer memory pools, Results

• Test 6: analysis of traffic protection for streams using the same buffer pool; Results