Test Description

1. Topology
2. Parameters:
   • EF frame size (it includes layer 2 overhead): [64, 1518] bytes
   • EF scheduling: WFQ and PQ
     8 queues in total are always used.
     With PQ, one queue is a priority queue (EF) while the remaining 7 are WFQ queues, each corresponding to a particular precedence value.
     In case of a WFQ-only system, the queue matching precedence 7 is considered the EF queue, where the departure rate is 3 times the maximum arrival rate, while the remaining 7 queues are treated as in the previous case, i.e. each of it corresponds to a precedence value and the departure rate equals the arrival rate.
   • matching: IP src and dst address and port-based
3. Stream profiles:
   • EF: load is 100 Kbps (constant rate), packet size variable, SmartBits
   • Background overall rate > 2.0 Mbps in each hop, real distribution
     Size contant for a given stream
     departure rate constantly exceeded
   • BE (prec 0): BE traffic produced by packets exceeding a configured rate at the ingress interface
   • prec 1 to 6 packets injected in the first router, prec0 and prec 7 traffic added in each hop to produce long-term congestion and aggregation
   • Script for generation correct this (first hop router)
   • prec 0 to 7 traffic: UDP
4. Router configuration
5. 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 minimizes delay in comparison with WFQ, whatever number of queues (2 to 8) is used;
• PQ minimizes jitter only if in WFQ several (e.g. 8) queues are active (as shown in this test, with just 2 queues jitter with PQ and WFQ is approx. equivalent)
• EF Jitter with PQ + 7 WFQ queues is constant (it does not change with the frame size).
• Delay and jitter frequency distribution functions are identically shaped with PQ - independently of the packet size -, while with WFQ for larger packet sizes the distribution gets more spread around the average.

Comments:

• ONE-WAY DELAY:
  Figure 1 plots average one-way delay with PQ and WFQ (8 queues). Delay with WFQ is always greater than the equivalent delay measured with PQ.
  See also the corresponding test with 2 queues test) for a comparison.
  The set of 3 graphs in Figure 2 compare the evolution over time of delay with PQ and WFQ for 3 different frame sizes: 128, 1024 and 1518 by.
  
  Fig.1: average one-way delay for different EF frame sizes, with PQ and WFQ (8 queues)
  Fig.2: comparison of one-way delay with PQ and WFQ with 128, 1024 and 1518 byte frame size
  One-way delay distribution with WFQ is identically shaped with 128 and 1518 byte frames, as illustrated in Figure 3 (a)
  Fig.3 (a): frequency delay distributions with PQ and WFQ (8 queues)
  Fig.3 (b): complementary cumulative density functions for one-way with WFQ and PQ for 128 and 1518 byte frames
• IPDV:
  Figure 4 shows that the difference in jitter is relevant in case of a 8-queue WFQ system. On the other hand, previous tests show that with only two queues jitter with PQ and WFQ is approximately equivalent (see this graph for a comparison).
  With PQ jitter is contant and this is also shown in graphs of Figure 5, which show how jitter varies over time for different EF frame sizes with PQ and WFQ.
  
  Fig.4: average IPDV with PQ and WFQ (8 queues)
  
  Fig.5: comparison of jitter with PQ and WFQ with 128, 1024 and 1518 by frame size
  
  Fig.6: jitter frequency distributions for 128 and 1518 by frames

  ---

  
  Pictures in large format:
  Fig. 3, Fig. 4, Fig. 5
  Fig. 6, Fig. 7, Fig. 8
  Fig. 9, Fig. 10
  Fig. 11, Fig. 12
  Fig. 13, Fig. 14 [end]