AFAIK hardware nat is only available for mt7621. This device has 1700mhz dual core CPU. Would that be sufficient. Last time I had ASUS RT-AC88U, which had 1400MHZ dual core, it wasn't enough really, for 1gbit. It would be 700Mbps at most, if I remember right. And %90 CPU usage while at that.
No, it won't.
You will find more detailed discussions about the performance of ipq8065 devices such as the nbg6817 or the r7800 via the forum search.
1 Like
Really? I've an Archer C6 v2 (US), which is a single core mips processor running at 775 mhz with no hardware flow offloading support; I can reach 600-650 Mbps wired with software flow offloading.
ASUS RT-AC88U packs an ARM processor, so it's difficult to imagine it being unable to reach 1 Gbps wired with software flow offloading, unless I am missing something.
Really.
(again, the reasons have been discussed in detail before, multiple times, so I won't repeat that here again)
Keep in mind, the CPU core itself and its processing power is not the whole story, its I/O components play a major part as well - as do the drivers. You can have a top of the line core i9, if you use '1 GBit/s' USB 2.0 ethernet cards, the results will still be bad - or if you put the graphics card into an x1 PCIe slot and want to play a 3d game. There is a reason why the filogic device will hit higher routing performance, despite having lower clocked CPU cores (and comparing mips against ARM SOCs is another topic altogether).
Use nss build, this is only one usefull openwrt, clean is useless.
Well, I was talking about his mention of not getting 1 gbps on a RT-AC88U, while CPU usage was peaking 90%.
I/O bottleneck would keep the CPU usage low, not maxed (or nearly maxed).
With that being said, it's a Broadcom SoC; I am surprised OpenWrt even runs on that.
Wrong, ipq8065 can handle gigabit NAT on latest OpenWRT master.
Here is a fresh iperf3 test, 80% CPU load on both cores.
SUM] 0.00-60.01 sec 6.42 GBytes 919 Mbits/sec 605536 sender
[SUM] 0.00-60.00 sec 6.38 GBytes 913 Mbits/sec receiver
It takes a few config changes though:
- Software flow offloading ON
- Packet steering ON
- Steering flows = 256
- IRQ affinities must be configured, add to /etc/rc.local
# eth0 wan
echo 1 > /proc/irq/32/smp_affinity_list
echo 3 > /sys/class/net/eth0/queues/tx-0/xps_cpus
# eth1 lan
echo 0 > /proc/irq/33/smp_affinity_list
echo 3 > /sys/class/net/eth1/queues/tx-0/xps_cpus
But generally, OpenWRT ipq806x had a rather turbulent history: performance fluctuating between releases, misconfigured defaults, unoptimized network drivers and CPU cache issues. WiFi ath10k part is very much hit and miss too: random firmware issues, ath10k vs ath10k-ct split, etc.
So avoid buying ipq806x unless you already have it or got is for pennies. Get Mediatek filogic instead.
Very nice! I've not upgraded a slightly slower clocked ipq8064 EA8500 donated to a neighbor from 23.05.3 out of concern conversion to DSA on main snapshot would be too much of a performance hit. It's encouraging to see your ipq8065 at least routing wired near line rate (with your config changes). It's probably not going to do that simultaneously managing WiFi, but within its limits with slower ISP service its not quite dead yet.
Agree completely with you and slh here - there are far better Mediatek ARM options today, even older MT7622, but certainly filogic.