Test Description

1. Parameters:
   • EF frame size (it includes layer 2 overhead): [64, 1518] bytes
   • EF scheduling: WFQ and PQ
2. Stream profiles:
   • EF: rate 300 Kbps (constant rate), packet size variable
   • BE: overall rate > 2.0 Mbps, multiple CBR streams, at different rates and with different packet size (the size is contant within a given stream)
     BE script
3. Test conditions:
   • EF queue-limit = 10 pack (constant)
   • tx-ring-limit: 5 particles
   • Priority Queuing policing rate = 300 Kbps
   • WFQ EF bandwidth = 300 Kbps
   • PVC: bandwidth 2 Mbps

Results in short:

• PQ is better hat WFQ in terms of one-way delay, in particular for relatively large packets (the difference is more relevant for large packet sizes). For small packets (64 bytes or less) the difference delays in the cases are equivalent.
• One-way delay with PQ for small packets is smaller, but more spread than with WFQ (PQ performance looks worse since one-way delay varies in a larger range).
  On the other hand with large packets, one-way delays are similarly distributed but in two ranges which do not intersect (PQ is definitely better than WFQ).
• The performace of the two schedulers is almost equivalent from the point of view of IPDV, both from the point of view of the sample mean and of the frequency distribution.

Comments:

• Figure 1 plots the one-way sample mean with both WFQ and PQ for different EF frame sizes.
  The gain with PQ is evident, and the difference between the two curves increases with the frame size, while for small packet sizes, WFQ converges to PQ.
• Figure 2 and 3 plot the one-way frequency distribution with PQ and WFQ for two different frame sizes: 128 bytes and 1518 bytes.
  With 128 bytes frames the sample mean is almost equivalent in the two cases, but the distribution is extremely different: with WFQ delay is more constant and samples are concentrated in a relatively small interval, while with PQ delay varies greatly. This can be explained by the fact that with PQ delay of a EF packet can vary greatly: if an EF packet is at the head of the queue it has to wait for the completion of the current BE transmission, but packets which arrive into the queue when a EF queue transmission is ongoing do not experience the queuing delay introduced by a BE packet.
  On the other hand, with the longer frame delay ranges with WFQ and PQ do not intersect, however, the frequency distribution looks pretty much the same.
• Figure 4 shows that PQ and WFQ have almost the same performance in terms of average IPDV. The equivalence of the algorithms is also evident from the IPDV frequency distributions in the two cases which were plotted for 128 by frames (Figure 5) and 1518 by frames (Figure 6).