I’ve pushed a new version of ethtscal that greatly improves the 
accuracy/stability of results. With runs of 50000 packets, I’m seeing 
consistency of +-5ns on loopback. If you pulled an older version, I recommend 
pulling a new one. 

Here is a sample run with i211 chips with loopback cable:

Iterating over 100000 packets (minimum runtime 200 seconds)

Tx interface igb0, speed 100Mb, adddress 0:8:a2:9:5b:8b
Rx interface igb1, speed 100Mb, adddress 0:8:a2:9:5b:8a

Tx avg ptp offset -200414826ns, 15 of 25 samples used
Rx avg ptp offset -96810099ns, 15 of 25 samples used

Misc times:
  Tx before send  ->  after send       8108ns
  Tx before send  ->  tx timestamp    15481ns
  Tx after send   ->  before poll    224578ns
  Tx after send   ->  after poll     231414ns
  Rx timestamp    ->  poll return     27296ns

Compensation values:
  Tx timestamp                            0ns
  Rx timestamp                            0ns
  Switch port to port                     0ns
  Cable length                            1ns  (0.3m)
  Total                                   1ns

Tx -> Rx timestamp:
  Uncompensated                        3175ns  (stddev 110ns)
  Compensated                          3175ns

Connection types:       Expected        Error
  Regular switch          6721ns      -3546ns
  Cut-through switch      2081ns       1094ns
  loopback cable             1ns       3174ns

Intel’s predicted result is 3177ns.

Denny


Reply via email to