Performance Tuning

Run 2 – review data – next steps

We moved interrupts and reran the test,
●
saw a very slight improvement in throughput,
●
not really expecting much...yet

Looking at the mpstat output we can see that we are still
bottlenecked on the receive side.

Let’s now try to add some process affinity to the mix
●
Remember core mappings
●
Try to get netperf to run on a “sister” core to the interrupts
●
Typically use taskset or TUNA for this, can also handle
programatically
●
netperf also has a built in mechanism