Test Description

1. Network layout
2. BE data streams:
   • multiple EF streams per site (20)
   • BE packet size: according to the real distribution
   • BE protocol: UDP
   • profile: CBR
3. Example of router configuration (with WFQ)
4. Parameters:
   • EF load: variable in the range [200, 1000] Kbps, i.e. [10, 50]% of line rate
   • EF IP payload packet size: 40 bytes or 512 bytes
   • EF scheduling algorithm: PQ or WFQ
5. Stream profiles:
   • 10 EF streams per site for 40 by packets (40 in total),
     1 EF stream per site for 512 by packets (10 in total)
     1 EF stream generated by the SmartBits
   • EF protocol: UDP, CBR
   • EF rate: same for each stream
   • examples of scripts per test site:
     • EF load 200 Kbps: 40 by and 512 by
     • EF load 400 Kbps: 40 by and 512 by
     • EF load 600 Kbps: 40 by and 512 by
     • EF load 800 Kbps: 40 by and 512 by
     • EF load: 1000 Kbps: 40 by and 512 by
6. Test conditions:
   • PVC: bandwidth 2 Mbps
   • queueing algorithm: priority queuing applied to EF
   • EF agregation and congestion in each transit site
   • 5000 samples per test
7. Test methodology
   We define EF laod the overall traffic volume produced by EF streams. The EF load varies in the range [200, 1000] Kbps and each site injects the same amount of traffic.
   For each test in each router on the data path we measure the amount of EF packets tail dropped by the priority queue (which serves EF packets) and we progressively increase the priority queue size Q until no packet loss caused by tail drop is observed during the whole test. Given the queue size Q at the time when such condition applies, we assume that the maximum burst size (in packets) is Q -1. The burst size in bytes can be derived since the EF packet size is constant and known.
   Measurement is applied to a single stream, which is called the reference stream.

   Two packet sizes: 40 by and 512 by, are deployed for WFQ testing, since for small EF packet sizes the WFQ scheduling algorithm converges to a PQ scheduler, while for larger packet sizes the behaviour of the schedulers differs.

Comments:

• Figure 1 compares the different profile of an EF stream in terms of burstiness with two different algorithms: WFQ and PQ. Burstiness with WFQ is measured for two EF IP payload sizes: 40 by and 512 by.
  The burstiness with PQ is measured when the EF payload size is equal to 40 bytes.

  As the figure shows, burstiness with WFQ and fairly large packet sizes is almost negligible, in fact, as expected, with WFQ the formation of large EF bursts is prevented, since even in case of packet clustering WFQ can interrupt the transmission of EF packets to send BE packets, this for its fairness.
  With shorter packet sizes (e.g. 40 by like in this test), WFQ converges to a PQ algorithm, since with WFQ short packets get a preferential treatment (the forwarding time computed by WFQ is proportional to the packet size). In this case, the test shows that burstiness with PQ and WFQ is very similar.

  Figure 1: EF burstiness with WFQ and PQ

  ---

  Last modified: Apr 10, 2000