Solved Irqbalance in relation to Cakeqos

[DJones]

I’ve asked about irqbalance in the past here. I understand the concern with broadcoms stuff. However I’m curious about it in regards to QoS-Cake since it does not use hardware acceleration. I’m curious if under that scenario would it help or manually assigning cores.

“Depends. If you need high performance PPPoE, the only way to make PPPoE become multithreaded with pfsense is actually to virtualize it, so NIC interrupts can be irq-balanced then. https://www.neelc.org/posts/opnsense-pppoe-kvm/” ~ @RMerlin

Under Linux according to the article at least under a bridge packets are forwarded without processing. But under QoS might they be processed multithreaded? I’m unsure if they already are or if it’s locked to core 0. I’m also kind of curious if pppoe performs better under multithreaded conditions in Linux or if that is only applicable to virtualized machines and because the topic quoted at hand was pfsense freebsd.

 

[sfx2000]

Let the kernel manage threads/processes - esp since you have PPPoE, which isi single threaded in any event...

 

[DJones]

Let the kernel manage threads/processes - esp since you have PPPoE, which isi single threaded in any event...

Thanks. I guess I should have prefaced this with it just being a general curiosity rather than actual technical support required. My connection plan isn’t fast enough for it to matter otherwise I wouldn’t even be using asuswrt-Merlin besides maybe as AP’s, and using pfsense as my main router.

I know a lot of things are blackboxed in asus products and mucking with configurations that may otherwise impact your router could cause regression in performance, errors, or worse.

 

[RMerlin]

Broadcom and Asus manually do affinity tuning at boot time. I trust Broadcom knows what's best there in terms of performance.

My router has no issue handling 1 Gbps over PPPoE with minimal CPU usage, so I suspect there might be some BCM blackbox magic at work there.

 

[DJones]

Thanks good enough to put my curiosity at bay.