Hi everyone, as described on the SQM wiki, I'm using my router with Software Stream Acceleration enabled and Packet Steering disabled. However, I'm wondering if activating IRQ Balancing while Software Stream Acceleration is on would result in even better performance?
Well, why don't you try it out and report your findings back to this thread?
2 Likes
Since the wiki is in English, I'm actually wondering if I can use IRQBalance with software flow acceleration enabled.
IMO, those work for some and not for others. The best practice is to discover your lan & wan device names, then do a cat /proc/interrupts.
Example for my Flint2 Running OpenWrt Stock v24.10.5:
CPU0 CPU1 CPU2 CPU3
11: 11711914 12565651 11767795 12595041 GICv3 30 Level arch_timer
24: 0 0 0 0 mt-eint 9 Edge keys
61: 41 0 0 0 mt-eint 46 Level mdio-bus:01
62: 2 0 0 0 mt-eint 47 Level mdio-bus:07
81: 24883 0 0 0 mt-eint 66 Level mt7530
116: 13 0 0 0 GICv3 155 Level ttyS0
120: 29683900 0 0 0 GICv3 229 Level 15100000.ethernet
121: 1129 74492954 0 0 GICv3 230 Level 15100000.ethernet
122: 0 0 0 0 GICv3 142 Level wdt_bark
123: 3581 0 0 0 GICv3 175 Level 11230000.mmc
124: 0 0 0 0 mt7530 0 Edge mt7530-0:00
125: 0 0 23787 0 mt7530 1 Edge mt7530-0:01
126: 0 0 0 0 mt7530 2 Edge mt7530-0:02
127: 0 0 1096 0 mt7530 3 Edge mt7530-0:03
128: 0 0 0 0 GICv3 205 Level xhci-hcd:usb1
129: 0 0 0 0 GICv3 148 Level 10320000.crypto
130: 0 0 0 0 GICv3 149 Level 10320000.crypto
131: 0 0 0 0 GICv3 150 Level 10320000.crypto
132: 0 0 0 0 GICv3 151 Level 10320000.crypto
133: 1944 0 0 85854898 GICv3 245 Level mt7915e
IPI0: 64573 346706 142982 105525 Rescheduling interrupts
IPI1: 44654909 101036256 91317788 36448343 Function call interrupts
IPI2: 0 0 0 0 CPU stop interrupts
IPI3: 0 0 0 0 CPU stop (for crash dump) interrupts
IPI4: 0 0 0 0 Timer broadcast interrupts
IPI5: 0 14529 10097 17457 IRQ work interrupts
IPI6: 0 0 0 0 CPU wake-up interrupts
Err: 0
/etc/rc.local
# Wait for interfaces to be fully up
sleep 10
# ---------------------------------------------------------
# FINALIZED CPU AFFINITY MAP
# Core 0 (1): System/Misc, LAN (eth0, br-lan, lan1-5)
# Core 1-2 (2,4): WAN (eth1, ifb4eth1)
# Core 3 (8): Wireless (when active)
# ---------------------------------------------------------
echo 1 > /proc/irq/120/smp_affinity
echo e > /sys/class/net/eth0/queues/rx-0/rps_cpus
echo e > /sys/class/net/br-lan/queues/rx-0/rps_cpus
for i in /sys/class/net/eth0/queues/tx-*/xps_cpus; do echo e > "$i"; done
for i in /sys/class/net/lan*/queues/rx-0/rps_cpus; do echo e > "$i"; done
echo 6 > /proc/irq/121/smp_affinity
echo e > /sys/class/net/eth1/queues/rx-0/rps_cpus
echo e > /sys/class/net/ifb4eth1/queues/rx-0/rps_cpus
for i in /sys/class/net/eth1/queues/tx-*/xps_cpus; do echo e > "$i"; done
echo 8 > /proc/irq/133/smp_affinity
for iface in phy0-ap0 phy1-ap0; do
[ -d "/sys/class/net/$iface" ] && echo e > /sys/class/net/$iface/queues/rx-0/rps_cpus 2>/dev/null
done
# Allow more packets per "tick" to compensate for the slow 100Hz timer
echo 1000 > /proc/sys/net/core/netdev_budget
# Increase the backlog so packets don't overflow between the slow ticks
echo 5000 > /proc/sys/net/core/netdev_max_backlog
exit 0
1. Dedicated CPU Lanes
Instead of letting all data pile up on one core, it divides the work:
- Core 0: Manages your Home Devices (LAN).
- Cores 1-2: Dedicated to the Internet (WAN) for max speed.
- Core 3: Reserved strictly for Wi-Fi stability.
2. Bigger "Storage Bins"
Because this router's internal clock is a bit slow, the script makes two vital changes:
- Larger Bites: It tells the CPU to process 1,000 packets at a time (instead of the small default) so it doesn't fall behind.
- Bigger Waiting Room: It increases the "Backlog" to 5,000, giving incoming data a place to sit if the CPU is momentarily busy.
The Result:
- Lower Ping: Great for gaming because traffic doesn't get stuck behind other tasks.
- No Stuttering: Prevents the router from "freezing" during heavy downloads.
- Better Wi-Fi: Wireless signals get their own dedicated processing power.
Mine are like this
CPU0 CPU1 CPU2 CPU3
8: 1990210 2135258 2339588 2656633 MIPS GIC Local 1 timer
9: 5523094 0 0 0 MIPS GIC 63 IPI call
10: 0 7087491 0 0 MIPS GIC 64 IPI call
11: 0 0 2737452 0 MIPS GIC 65 IPI call
12: 0 0 0 1500581 MIPS GIC 66 IPI call
13: 11561 0 0 0 MIPS GIC 67 IPI resched
14: 0 20545 0 0 MIPS GIC 68 IPI resched
15: 0 0 3805 0 MIPS GIC 69 IPI resched
16: 0 0 0 3395 MIPS GIC 70 IPI resched
17: 0 0 0 0 MIPS GIC 19 1e000600.gpio-bank0, 1e000600.gpio-bank1, 1e000600.gpio-bank2
18: 12 0 0 0 MIPS GIC 33 ttyS0
19: 5440897 0 0 0 MIPS GIC 10 1e100000.ethernet
20: 0 0 0 0 MIPS GIC 30 mt7530
23: 0 0 0 0 mt7530 2 mt7530-0:02
24: 0 0 0 0 mt7530 3 mt7530-0:03
25: 0 0 0 0 mt7621-gpio 18 keys
26: 0 0 0 0 MIPS GIC 26 1e004000.crypto
27: 23 0 0 0 MIPS GIC 31 mt7603e
28: 960 0 0 14096115 MIPS GIC 11 mt76x2e
ERR: 51150
You have LOTS of errors. run this:
for i in /proc/irq/*/smp_affinity_list; do echo -n "IRQ ${i#*/irq/}: "; cat $i; done
1 Like
Do you use irqbalance and packet_steering Package in the background!!
For irq for my wan i use vlan eth1.100, do i have to change
No. I have both either disabled/not installed. To each their own but I find them to be very inconsistent on various SOCs.
run this and we can see what you have:
grep . /proc/irq/*/smp_affinity
for i in /proc/irq/*/smp_affinity_list; do echo -n "IRQ ${i#*/irq/}: "; cat $i; done
ls /sys/class/net/eth*/queues/tx-*/xps_cpus | xargs cat
ls -l /sys/class/net/*/device
uci get network.wan.device
The basic idea is keep everything that isn't LAN & system related such as WAN, WiFi, etc off of eth0 but on dedicated processors, since modern systems have more than 2 cores. RPS should spread to everything but eth0. We add the hex numbers together for more than a single cpu...
| Column 1 | Column 2 | Column 3 | Column 4 |
|---|---|---|---|
| Hex | Binary | Decimal | |
| 0 | 0 | 0 | |
| 1 | 1 | 1 | |
| 2 | 10 | 2 | |
| 3 | 11 | 3 | |
| 4 | 100 | 4 | |
| 5 | 101 | 5 | |
| 6 | 110 | 6 | |
| 7 | 111 | 7 | |
| 8 | 1000 | 8 | |
| 9 | 1001 | 9 | |
| A | 1010 | 10 | |
| B | 1011 | 11 | |
| C | 1100 | 12 | |
| D | 1101 | 13 | |
| E | 1110 | 14 | |
| F | 1111 | 15 | |
| 14 | 10100 | 20 | |
| 3F | 111111 | 63 |
The Example Math
The Hex e comes from a "bitmask" representing 4 CPU cores:
- Core 0 = 1
- Core 1 = 2
- Core 2 = 4
- Core 3 = 8
2+4+8=14 decimal
Converted to hex= e
1 Like
root@OpenWrt:~# for i in /proc/irq/*/smp_affinity_list; do
> echo -n "IRQ ${i##*/irq/}: "
> cat $i
> done
IRQ 0/smp_affinity_list: 0-3
IRQ 1/smp_affinity_list: 0-3
IRQ 10/smp_affinity_list: 0-3
IRQ 11/smp_affinity_list: 0-3
IRQ 12/smp_affinity_list: 0-3
IRQ 13/smp_affinity_list: 0-3
IRQ 14/smp_affinity_list: 0-3
IRQ 15/smp_affinity_list: 0-3
IRQ 16/smp_affinity_list: 0-3
IRQ 17/smp_affinity_list: 0-3
IRQ 18/smp_affinity_list: 0-3
IRQ 19/smp_affinity_list: 0-3
IRQ 2/smp_affinity_list: 0-3
IRQ 20/smp_affinity_list: 0-3
IRQ 23/smp_affinity_list: 0-3
IRQ 24/smp_affinity_list: 0-3
IRQ 25/smp_affinity_list: 0-3
IRQ 26/smp_affinity_list: 0-3
IRQ 27/smp_affinity_list: 0-3
IRQ 28/smp_affinity_list: 0-3
IRQ 3/smp_affinity_list: 0-3
IRQ 4/smp_affinity_list: 0-3
IRQ 5/smp_affinity_list: 0-3
IRQ 6/smp_affinity_list: 0-3
IRQ 7/smp_affinity_list: 0-3
IRQ 8/smp_affinity_list: 0-3
IRQ 9/smp_affinity_list: 0-3
root@OpenWrt:~#
Last login: Tue Jan 20 14:18:42 on ttys000
arda@Arda-iMac ~ % ssh root@192.168.1.1
BusyBox v1.36.1 (2025-12-30 22:05:19 UTC) built-in shell (ash)
_______ ________ __
| |.-----.-----.-----.| | | |.----.| |_
| - || _ | -__| || | | || _|| _|
|_______|| __|_____|__|__||________||__| |____|
|__| W I R E L E S S F R E E D O M
-----------------------------------------------------
OpenWrt 24.10.5, r29087-d9c5716d1d
-----------------------------------------------------
=== WARNING! =====================================
There is no root password defined on this device!
Use the "passwd" command to set up a new password
in order to prevent unauthorized SSH logins.
--------------------------------------------------
root@OpenWrt:~# grep . /proc/irq/*/smp_affinity
/proc/irq/0/smp_affinity:f
/proc/irq/1/smp_affinity:f
/proc/irq/10/smp_affinity:f
/proc/irq/11/smp_affinity:f
/proc/irq/12/smp_affinity:f
/proc/irq/13/smp_affinity:f
/proc/irq/14/smp_affinity:f
/proc/irq/15/smp_affinity:f
/proc/irq/16/smp_affinity:f
/proc/irq/17/smp_affinity:f
/proc/irq/18/smp_affinity:f
/proc/irq/19/smp_affinity:f
/proc/irq/2/smp_affinity:f
/proc/irq/20/smp_affinity:f
/proc/irq/23/smp_affinity:f
/proc/irq/24/smp_affinity:f
/proc/irq/25/smp_affinity:f
/proc/irq/26/smp_affinity:f
/proc/irq/27/smp_affinity:f
/proc/irq/28/smp_affinity:f
/proc/irq/3/smp_affinity:f
/proc/irq/4/smp_affinity:f
/proc/irq/5/smp_affinity:f
/proc/irq/6/smp_affinity:f
/proc/irq/7/smp_affinity:f
/proc/irq/8/smp_affinity:f
/proc/irq/9/smp_affinity:f
root@OpenWrt:~# for i in /proc/irq/*/smp_affinity_list; do echo -n "IRQ ${i##*/i
rq/}: "; cat $i; done
IRQ 0/smp_affinity_list: 0-3
IRQ 1/smp_affinity_list: 0-3
IRQ 10/smp_affinity_list: 0-3
IRQ 11/smp_affinity_list: 0-3
IRQ 12/smp_affinity_list: 0-3
IRQ 13/smp_affinity_list: 0-3
IRQ 14/smp_affinity_list: 0-3
IRQ 15/smp_affinity_list: 0-3
IRQ 16/smp_affinity_list: 0-3
IRQ 17/smp_affinity_list: 0-3
IRQ 18/smp_affinity_list: 0-3
IRQ 19/smp_affinity_list: 0-3
IRQ 2/smp_affinity_list: 0-3
IRQ 20/smp_affinity_list: 0-3
IRQ 23/smp_affinity_list: 0-3
IRQ 24/smp_affinity_list: 0-3
IRQ 25/smp_affinity_list: 0-3
IRQ 26/smp_affinity_list: 0-3
IRQ 27/smp_affinity_list: 0-3
IRQ 28/smp_affinity_list: 0-3
IRQ 3/smp_affinity_list: 0-3
IRQ 4/smp_affinity_list: 0-3
IRQ 5/smp_affinity_list: 0-3
IRQ 6/smp_affinity_list: 0-3
IRQ 7/smp_affinity_list: 0-3
IRQ 8/smp_affinity_list: 0-3
IRQ 9/smp_affinity_list: 0-3
root@OpenWrt:~# ls /sys/class/net/*/queues/tx-*/xps_cpus | xargs cat
cat: read error: No such file or directory
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
cat: read error: No such file or directory
cat: read error: No such file or directory
cat: read error: No such file or directory
cat: read error: No such file or directory
cat: read error: No such file or directory
cat: read error: No such file or directory
cat: read error: No such file or directory
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
root@OpenWrt:~# ls -l /sys/class/net/*/device
lrwxrwxrwx 1 root root 0 Jan 19 23:33 /sys/class/net/eth0/device -> ../../../1e100000.ethernet
lrwxrwxrwx 1 root root 0 Jan 19 23:33 /sys/class/net/lan1/device -> ../../../mdio-bus:1f
lrwxrwxrwx 1 root root 0 Jan 19 23:33 /sys/class/net/lan2/device -> ../../../mdio-bus:1f
lrwxrwxrwx 1 root root 0 Jan 19 23:33 /sys/class/net/phy1-ap0/device -> ../../../0000:01:00.0
lrwxrwxrwx 1 root root 0 Jan 19 23:33 /sys/class/net/wan/device -> ../../../1e100000.ethernet
root@OpenWrt:~# uci get network.wan.device
wan.35
root@OpenWrt:~#
I'm booting my router with this command, and my Ethernet connection is now splitting between CPU0 and CPU1. Is this correct?
sleep 5; echo e > /proc/irq/19/smp_affinity; [ -f /sys/class/net/eth0/queues/rx-0/rps_cpus ] && echo e > /sys/class/net/eth0/queues/rx-0/rps_cpus; for f in /sys/class/net/eth0/queues/tx-*/xps_cpus; do [ -f "$f" ] && echo e > "$f"; done; echo 32768 > /proc/sys/net/core/netdev_max_backlog
I use the same router Flint2 but im running openwrt ver. 24.10.3
root@OpenWrt:~# grep . /proc/irq/*/smp_affinity
/proc/irq/1/smp_affinity:f
/proc/irq/10/smp_affinity:f
/proc/irq/11/smp_affinity:f
/proc/irq/116/smp_affinity:f
/proc/irq/12/smp_affinity:f
/proc/irq/120/smp_affinity:1
/proc/irq/121/smp_affinity:6
/proc/irq/122/smp_affinity:f
/proc/irq/123/smp_affinity:f
/proc/irq/124/smp_affinity:f
/proc/irq/125/smp_affinity:f
/proc/irq/126/smp_affinity:f
/proc/irq/127/smp_affinity:f
/proc/irq/128/smp_affinity:f
/proc/irq/129/smp_affinity:1
/proc/irq/13/smp_affinity:f
/proc/irq/130/smp_affinity:2
/proc/irq/131/smp_affinity:4
/proc/irq/132/smp_affinity:8
/proc/irq/133/smp_affinity:8
/proc/irq/2/smp_affinity:f
/proc/irq/24/smp_affinity:f
/proc/irq/3/smp_affinity:f
/proc/irq/4/smp_affinity:f
/proc/irq/5/smp_affinity:f
/proc/irq/6/smp_affinity:f
/proc/irq/61/smp_affinity:f
/proc/irq/62/smp_affinity:f
/proc/irq/7/smp_affinity:f
/proc/irq/8/smp_affinity:f
/proc/irq/81/smp_affinity:f
/proc/irq/9/smp_affinity:f
root@OpenWrt:~# for i in /proc/irq/*/smp_affinity_list; do echo -n "IRQ ${i#*/irq/}: "; cat $i; done
IRQ 1/smp_affinity_list: 0-3
IRQ 10/smp_affinity_list: 0-3
IRQ 11/smp_affinity_list: 0-3
IRQ 116/smp_affinity_list: 0-3
IRQ 12/smp_affinity_list: 0-3
IRQ 120/smp_affinity_list: 0
IRQ 121/smp_affinity_list: 1-2
IRQ 122/smp_affinity_list: 0-3
IRQ 123/smp_affinity_list: 0-3
IRQ 124/smp_affinity_list: 0-3
IRQ 125/smp_affinity_list: 0-3
IRQ 126/smp_affinity_list: 0-3
IRQ 127/smp_affinity_list: 0-3
IRQ 128/smp_affinity_list: 0-3
IRQ 129/smp_affinity_list: 0
IRQ 13/smp_affinity_list: 0-3
IRQ 130/smp_affinity_list: 1
IRQ 131/smp_affinity_list: 2
IRQ 132/smp_affinity_list: 3
IRQ 133/smp_affinity_list: 3
IRQ 2/smp_affinity_list: 0-3
IRQ 24/smp_affinity_list: 0-3
IRQ 3/smp_affinity_list: 0-3
IRQ 4/smp_affinity_list: 0-3
IRQ 5/smp_affinity_list: 0-3
IRQ 6/smp_affinity_list: 0-3
IRQ 61/smp_affinity_list: 0-3
IRQ 62/smp_affinity_list: 0-3
IRQ 7/smp_affinity_list: 0-3
IRQ 8/smp_affinity_list: 0-3
IRQ 81/smp_affinity_list: 0-3
IRQ 9/smp_affinity_list: 0-3
root@OpenWrt:~# ls /sys/class/net/eth*/queues/tx-*/xps_cpus | xargs cat
e
e
e
e
e
e
e
e
e
e
e
e
e
e
e
e
cat: read error: No such file or directory
e
e
e
e
e
e
e
e
e
e
e
e
e
e
e
e
root@OpenWrt:~# ls -l /sys/class/net/*/device
lrwxrwxrwx 1 root root 0 Jan 19 23:04 /sys/class/net/eth0/device -> ../../../15100000.ethernet
lrwxrwxrwx 1 root root 0 Jan 19 23:04 /sys/class/net/eth1/device -> ../../../15100000.ethernet
lrwxrwxrwx 1 root root 0 Jan 19 23:04 /sys/class/net/lan1/device -> ../../../mdio-bus:1f
lrwxrwxrwx 1 root root 0 Jan 19 23:04 /sys/class/net/lan2/device -> ../../../mdio-bus:1f
lrwxrwxrwx 1 root root 0 Jan 19 23:04 /sys/class/net/lan3/device -> ../../../mdio-bus:1f
lrwxrwxrwx 1 root root 0 Jan 19 23:04 /sys/class/net/lan4/device -> ../../../mdio-bus:1f
lrwxrwxrwx 1 root root 0 Jan 19 23:04 /sys/class/net/lan5/device -> ../../../mdio-bus:1f
root@OpenWrt:~# uci get network.wan.device
eth1.100
How many total cores do you have?
please paste the output of:
cat /proc/interrupts
e means to use cores 1-3, NOT 0.
GL.iNet GL-MT6000 have 4 cores
root@OpenWrt:~# cat /proc/interrupts
CPU0 CPU1 CPU2 CPU3
11: 260196 963626 1142905 689929 GICv3 30 Level arch_timer
24: 0 0 0 0 mt-eint 9 Edge keys
61: 13 0 0 0 mt-eint 46 Level mdio-bus:01
62: 2 0 0 0 mt-eint 47 Level mdio-bus:07
81: 6 0 0 0 mt-eint 66 Level mt7530
116: 13 0 0 0 GICv3 155 Level ttyS0
120: 1115701 0 0 0 GICv3 229 Level 15100000.ethernet
121: 2825 1126862 0 0 GICv3 230 Level 15100000.ethernet
122: 0 0 0 0 GICv3 142 Level wdt_bark
123: 2262 0 0 0 GICv3 175 Level 11230000.mmc
124: 3 0 0 0 mt7530 0 Edge mt7530-0:00
125: 3 0 0 0 mt7530 1 Edge mt7530-0:01
126: 0 0 0 0 mt7530 2 Edge mt7530-0:02
127: 0 0 0 0 mt7530 3 Edge mt7530-0:03
128: 0 0 0 0 GICv3 205 Level xhci-hcd:usb1
129: 0 0 0 0 GICv3 148 Level 10320000.crypto
130: 0 0 0 0 GICv3 149 Level 10320000.crypto
131: 0 0 0 0 GICv3 150 Level 10320000.crypto
132: 0 0 0 0 GICv3 151 Level 10320000.crypto
133: 744 0 0 0 GICv3 245 Level mt7915e
IPI0: 7980 7731 8549 8374 Rescheduling interrupts
IPI1: 276988 779298 975287 687185 Function call interrupts
IPI2: 0 0 0 0 CPU stop interrupts
IPI3: 0 0 0 0 CPU stop (for crash dump) interrupts
IPI4: 0 0 0 0 Timer broadcast interrupts
IPI5: 0 0 0 0 IRQ work interrupts
IPI6: 0 0 0 0 CPU wake-up interrupts
Err: 0
Breakdown of Operations
1. Wait Period
sleep 10: Ensures network interfaces are fully initialized before applying hardware-level configurations.
2. Core 0: System & LAN Management
- IRQ 120 (eth0): Bound to Core 0 (
affinity 1). - RPS/XPS: Configures Receive Packet Steering (RPS) and Transmit Packet Steering (XPS) for
eth0,br-lan, and individuallan*ports. - Mask
e: Despite the comment suggesting Core 0, the maske(binary1110) actually delegates software processing to Cores 1, 2, and 3, keeping Core 0 free for hardware interrupts and system tasks.
3. Cores 1–2: WAN Traffic
- IRQ 121 (eth1): Bound to Cores 1 and 2 (
affinity 6, binary0110). - Traffic Shaping: Applies RPS/XPS to
eth1and its associated Intermediate Functional Block (ifb4eth1), which is typically used for bufferbloat control (SQM).
4. Core 3: Wireless Operations
- IRQ 133 (WiFi): Bound to Core 3 (
affinity 8, binary1000). - Wireless Interfaces: Directs software packet steering for
phy0-ap0andphy1-ap0(2.4GHz and 5GHz radios) to this core to prevent WiFi processing from lagging the wired WAN/LAN.
5. Kernel Tuning for Low-HZ Systems
netdev_budget(1000): Increases the number of packets the CPU can process in one interrupt "poll" cycle.netdev_max_backlog(5000): Increases the queue size for packets waiting to be processed by the CPU, preventing drops during sudden traffic spikes.
This topic was automatically closed 10 days after the last reply. New replies are no longer allowed.