Instantaneous packet delay variation over time
Time is represented by the packet sequence number. The assumption is that the traffic is issued by a constant bit rate source (300 Kbps) and that the inter packet gap is constant, so that for each time interval only a single packet is received.

Comments:
Note: noBE: without best-effort, BE: with best-effort

  1. As for one-way delay the PBS does not impact the ipdv metric, both with anf without best-effort traffic. This is true when the PBS is one of the following: [5000, 10000, 15000, 20000, 45000].
    While ipdv is the same for anyPBS value when no best-effort traffic is injected, PBS = 1500 with best-effort makes ipdv increase considerably. This behaviour could be due to implementation problems; this is under investigation.
  2. (noBE) Figure 13 and 14 show the ipdv metric for several packet sizes: 64, 256, 1024 and 1518 bytes.
    We see for a given packet size ipdv oscillates around a given mean value which gradually increases up to 1.09 msec and then it keeps constant.
    However, the magnitude of the oscillation changes irregularly with the packet size, in the sense that it is almost constant for packet sizes of 64 bytes and 1518 bytes, while it changes a lot and periodically for packets of 1024 bytes.
    For packet sizes of 64 bytes figure 13 shows some ipdv peaks: We can assume that they are due to asynchronous events due to the existence of concurrent processes in the router, since on the IBM 2212 scheduling is implemented in software.
  3. (noBE) ipdv is the same for any Pbs value (but only without best-effort traffic).
  4. The following table (extracted from figure 13, 14, 15, 16 and 17) compares the ipdv range values for different test scenarios. 
    pack size   test type	PBS	range 			delta
(bytes)			(bytes)	(msec)			(msec)

    
64         noBE		1500	[ 0.0109, 0.0166 ](*)	0.0057
256        noBE		"	[ 0.6975, 1.5146 ]	0.8171
1024       noBE		"	[ 0.9723, 1.2410 ]	0.2687
1518       noBE		"	[ 1.0929, 1.1192 ]	0.0263

64         BE		"	[ 0.0155, 10.1025]	10.087
256        BE		"	[ 0.0344, 7.3379 ]	7.3035
1024       BE		"	[ 1.1549, 7.7657 ]	6.6108
1518       BE		"	[ 0.8809, 8.0348 ]	7.1539

64         BE		10000	[ 0.0023, 4.5553 ]	4.5530
256        BE		"	[ 0.6727, 3.7988 ]	3.1261
1024       BE		"	[ 1.0623, 3.3030 ]	2.2407
1518       BE		"	[ 0.8846, 3.5806 ]	2.6960

(*) instantaneous peaks are not considered in this range
    We can see that ipdv oscillation ranges increases greatly when best-effort traffic is added. For PBS equal to 1500 bytes, the increase is well above the tranmission time of a best-effort packet (1000 bytes). however, the increase in ipdv is much bigger than the transmission time of a 1000 bytes best-effort traffic, so for PBS=1500 bytes we could assume that an implementation problem occurs. This is subject of further investigation.

    In comparison with PHB=1500 by, the increase in ipdv with best-effort for PHB=10000 bytes is much more limited (table above and figure 17) and is in the order of magnitude of 1 best-effort packet transmission time.

Figure 13: ipdv without best-effort traffic for packet sizes of 64 and 256 bytes (PBS = 1500 bytes).
Figure 14: idvp without best-effort traffic for packet sizes of 1024 and 1518 bytes.
Figure 15: same as figure 13 but with best-effort traffic. The maximum ipdv value greatly increases for both packet sizes.
Figure 16: same as figure 14 but with best-effort traffic. The maximum ipdv value greatly increases for both packet sizes.
Figure 17: same as figure 13 but with PBS of 10000 bytes. When best-effort traffic is added the ipdv increase is much more limited than with PBS= 1500 bytes (figure 15).
Last modified: Nov 22, 1999