why irqbalance balances interrupts on all numa nodes

Latest response

I have machine (Fujitsu RX300 S7) which has two numa nodes (2x6core XEON CPU). There is dual 10Gbit NIC connected to CPU1.

I can see via /proc/interrupts, that NIC (detected 24 IRQs - HT enabled) has balanced interrupts on all 24 cores on both NUMA nodes.

IRQBALANCE should balance interrupts only on local NUMA node, where NIC is connected (cores 0-5 and 12-17 which is CPU1). This is the same for RHEL7 and RHEL8.

Does someone have idea what is wrong?



(sorry for my earlier incorrect response if you got the email, I misunderstood the inverse of this)

Interesting, it's possible the hardware platform or device is not reporting NUMA correctly in a way that the kernel or irqbalance understands.

You could open a support case to investigate this in deeper detail.