Open ramaza opened 6 years ago
Your program should be very similar to bridge
, so maybe you should see if the same happens with bridge
. Also, you should add timestamps at poll
/ioctl
to check where the delay is introduced (input from VLAN A, input from VLAN B, output to VLAN A, output to VLAN B.
I'm testing a pretty simple zerocopy Netmap application that is bridging packets between two VLANs. I'm experiencing an interesting phenomenon in the results of a basic test with 1 ping per second.
The roundtrip-time is steadily shrinking from a pretty high level of 10ms to a rather normal level of less than 1ms over a period of about 100 seconds. Then it's jumping back to 10ms and the cycle starts again. There's no packet loss or any other problem. Also the throughput under high load is looking good. It's just this periodic sawtooth pattern in the roundtrip-time that I can't stop thinking about.
It reminds me a little bit about the superimposition of two timers with slightly different frequency. Maybe it's the ping interval in combination with some NAPI polling timer?
I'm thankful for any idea to narrow down this problem.