Since vsperf (ixnet trafficgen) started using AggResults.csv to pull out traffic results after bi-dir support was added. Some of the derived results are wrong (derived meaning these are figures vsperf calculates from the info available in the raw ixnet results). For example here for perf/continuous the tx_fps is lower than the rx_fps:

... throughput_tx_fps, Value: 6439197.
... throughput_rx_fps, Value: 13525168.014
...

Further Analysis

Clearly all we can get from AggregateResults.csv is:

Tx Rate (% Line Rate)
Rx Throughput (% Line Rate)
Rx Throughput (fps)
<Ignore latency figures - not to be an issue>

Given that:

• some tests (vlan frame mod) may change the packet sizes as they pass through the DUT
• some tests involved packet loss (simple perf test (MaxFowardingRate) - NOTE forwarding rate (lossy) is different to throughput (lossless).

Is there any more of the currently reported results fields:

• throughput_tx_fps,
• throughput_rx_fps,
• throughput_tx_mbps,
• throughput_rx_mbps,
• throughput_tx_percent,
• throughput_rx_percent,

that we can calculate? (I don't think so).

What if we look at using the other results .csv files? Where the traffic is broken down by input/output. Can we 'fill-in-the-blanks' then?

Tasks

• Establish what results we can correctly derive from only AggResults.csv
• Estimate the effort required to implement code to derive or extract all the currently stated results - this will require parsing additional .csv files from ixnet.
• Decide what extra figures if any are required and implement the derivation/extraction of these.

DOD

• All results from ixnet traffic generator are correct in all cases for all tests - bidi, unidir, frame modification
• Code is upstreamed to opnfv