We are using one of our T2080RDB (running SDK-1.9 / linux-3.12.37) to receive a continous stream of 9k jumbo packets (UDP) on the 10G link (fman1-mac9).
Compared to SDK-1.6 where packets were lost, SD-1.9 behaves a lot better.
However we are experiencing the following problems:
- UDP packets are received out-of-order (using recvfrom in userspace).
In /proc/interrupts it can be seen that QMan-Portal 0 -> 7 are active and handle interrupts each on a dedicated core, so maybe distributing the packets to multiple QMan portals could be an issue?
- The system experiences a quite high software interrupt rate during this kind of load, top reports:Cpu(s): 2.0%us, 0.8%sy, 0.0%ni, 48.2%id, 0.0%wa, 1.8%hi, 47.2%si, 0.0%st
Is there any way how we can receive the packets in order?
Are there tuning possibilities for our kind of workloads to get the software interrupt rate down?
Thank you in advance, Clemens