Tnx @Borromini tried your build yesterday. Running smooth! What do I need to config to get DHCP ipv4 up and running on lan ? Looks like it doesnt serve ipv4 addresses on default ? Or start with a clean config ?
Looks like this build is using unbound + odhcpd instead of dnsmasq + odhcpd-ipv6only, sorry. I can put up another build later on with OpenWrt defaults, or you can consult the wiki on how to set up the unbound/odhcpd combo.
check. will do some reading on unbound + odhcpd first for now 
hi
there is also a depandency problem on this firmware
kernel not compatible
example to download kmod-tcp-bbr
As the download page states, the build comes as-is. Read up on how OpenWrt uses kernel hashes to force you to install 'matching' kmods to a running kernel.
As long as RB5009UG support doesn't get merged, you'll have to help yourself. If you have a special use case it's very much worth it to invest the time in learning the buildroot, instead of having to rely on someone else's images.
1 Like
yes thanks for the response
i has created also my own firmware with ubuntu
1 Like
Image has been replaced with one with the default dnsmasq/odhcpd-ipv6only combo.
any news about support hsgmii? so sfp+ at 2.5gbit and lan too?
37 pieces in stock at Getic if anyone needs one:
Tnx! Currently running my own build based on 22.03 and your included manifest. Curious; you include a few iptables-mod packages in your build. Any reason for that as 22.03 uses nftables as I understand ?
1 Like
Pulled in by the likes of SQM or other stuff still depending on iptables, or leftovers carried over from when OpenWrt was still on iptables, probably.
1 Like
I was research some other thing but came across this thread from 2021, has this patch been added into the testing tree? This may fix the 2.5G port?
The issue seems specific to the QCA part, that patch set you linked to is already part of the RB5009UG patch set from what I can see.
1 Like
I recently got my hands on this router, because it looked like a cool board to play with, but experienced rather poor performance, considering it's a fairly modern 1.4Ghz quad core.
Because I use cake SQM with all the bells and whistles on two load-balanced connections, I'm used to seeing fairly high CPU load, but this new 1.4Ghz A72 Cortex was showing worse performance than my aging WRT1900ACS. Download speeds were very inconsistent (I load balance a 100Mbit + 500Mbit connection), hovering between 200-400Mbit, increasing and decreasing during various online speedtests.
Taking a look at htop showed CPU usage a little under 70% typically, switching to fq_codel improved speeds and CPU usage. After installing netdata I noticed the CPU frequency was not behaving as I expected. Averaging out to about 1Ghz max during large downloads or flent rrul tests on the busiest core cluster (the RB5009 has 2 core clusters of 2 cores each, with linked frequencies inside a cluster) and 700-ish Mhz on the less busy cluster.
After switching to the performance governor instead of the normal ondemand governor, performance was much improved, maxing out the connection.
echo performance > /sys/devices/system/cpu/cpufreq/policy0/scaling_governor
echo performance > /sys/devices/system/cpu/cpufreq/policy2/scaling_governor
Changing back to the ondemand governor and aggressively tuning its settings showed similar performance, with still some powersaving at low loads:
echo ondemand > /sys/devices/system/cpu/cpufreq/policy0/scaling_governor
echo ondemand > /sys/devices/system/cpu/cpufreq/policy2/scaling_governor
echo 10 > /sys/devices/system/cpu/cpufreq/ondemand/up_threshold
echo 10 > /sys/devices/system/cpu/cpufreq/ondemand/sampling_down_factor
7 Likes
Did you get a chance to monitor power consumption between performance governor, your tuned ondemand settings and the default ones? Curious about that.
Was in a rush so only checked idle usage, but it made no difference switching between stock, performance or the tuned ondemand in terms of power usage.
All settings hovered between 8.3 and 9.5 watts, measured at the wall with one of those plug meters.
Will try during a flent test later.
1 Like
Power readings under load (flent rrul):
Stock: 10.2W
Tuned ondemand: 10.3W
Performance: 10.3W
These numbers are very consistent during the time the flent rrul test is downloading and uploading, but at stock, I'm unable to fully max out the connection. Straight after the flent rrul test, they drop back to an idle 8.8W for tuned ondemand and performance, and 8.7W for the stock settings.
I think it makes little sense to use anything other than the performance governor.
2 Likes
Has anyone given 5.15 (which is the testing kernel for mvebu) a shot already, combined with adrons' patches? I just did a sysupgrade on a 'spare' RB5009UG from 5.10 to 5.15; it's not coming up anymore 
.
Same thing happened here. Tried applying the patches for 5.10 to 5.15, but that turned into a mess of merging 5.10 specific stuff with 5.15 which already contains some of those things and some files changed considerably in 5.15.
1 Like
I didn't even think of that. I just ticked 'use testing kernel' and YOLO 
. Bad idea 
.
I do know a lot of the Marvell switch code was backported from 5.15 to the 5.10 build by adrons. I have been looking a bit at it before (because of the QCA8081 NIC not working at 2,5 Gbps) and what's in the 5.10 build is basically the 5.15 codebase for that Marvell driver.