SvenRoederer
commented
3 years ago This a dump od /proc/interrupts" from a ubnt ERX after some days uptime. There is only the ethernet-driver which is relevant for pinning. So these boards will likely not qualify for irq-pinning.
root@Taubenschlag:~# cat /proc/interrupts
CPU0 CPU1 CPU2 CPU3
8: 104453730 104453783 104453821 104453825 MIPS GIC Local 1 timer
9: 113055 0 0 0 MIPS GIC 63 IPI call
10: 0 3274564 0 0 MIPS GIC 64 IPI call
11: 0 0 1291976 0 MIPS GIC 65 IPI call
12: 0 0 0 866971 MIPS GIC 66 IPI call
13: 1926360 0 0 0 MIPS GIC 67 IPI resched
14: 0 5108401 0 0 MIPS GIC 68 IPI resched
15: 0 0 1014618 0 MIPS GIC 69 IPI resched
16: 0 0 0 737050 MIPS GIC 70 IPI resched
17: 0 0 0 0 MIPS GIC 19 1e000600.gpio-bank0, 1e000600.gpio-bank1, 1e000600.gpio-bank2
19: 12 0 0 0 MIPS GIC 33 ttyS0
20: 14013701 0 0 0 MIPS GIC 10 1e100000.ethernet
22: 0 0 0 0 1e000600.gpio 12 keys
ERR: 0In contrast a Mikrotik RB750Gr3, there is in addition the USB and SDCard. For this board irq-pinning should improve, at least in theory.
root@Verklaerung-core:~# cat /proc/interrupts
CPU0 CPU1 CPU2 CPU3
8: 232833352 232833334 232833326 232833319 MIPS GIC Local 1 timer
9: 2081009 0 0 0 MIPS GIC 63 IPI call
10: 0 6174287 0 0 MIPS GIC 64 IPI call
11: 0 0 34501949 0 MIPS GIC 65 IPI call
12: 0 0 0 8520251 MIPS GIC 66 IPI call
13: 11782796 0 0 0 MIPS GIC 67 IPI resched
14: 0 13225794 0 0 MIPS GIC 68 IPI resched
15: 0 0 9478747 0 MIPS GIC 69 IPI resched
16: 0 0 0 4671113 MIPS GIC 70 IPI resched
19: 12 0 0 0 MIPS GIC 33 ttyS0
20: 16462 0 0 0 MIPS GIC 27 1e130000.sdhci
21: 0 0 0 0 MIPS GIC 29 xhci-hcd:usb1
22: 119039977 0 0 0 MIPS GIC 10 1e100000.ethernet
23: 9 0 0 0 MIPS GIC 30 gsw
ERR: 1
with the growing number of multi-CPU boards around it might be usefull to ensure that not a single CPU is doing all the work, while others are sleeping only. A pinning of interrupt to specific CPUs can be done manually, via /proc/ in the kernel. To automate this the package "irqbalance" exists. Reading the documents comes to the point the such pinning will only give a benefit for Multi-CPU systems having multiple IRQ-sources (e.g. ethernet(s), WiFi(s).
As I don't have such boards in reach, I can't tell which performance improvement this will give. I expect that the community can provide some data (/proc/interrupts, throughput-meassurements).
In addition the irqbalance daemon is not activated during package-install, which might require to create a "freifunk-berlin-smp-system" package to do the UCI-change, depend on irqbalance and probably do board specific changes.