But somehow i'm struggling to get to 1gbps with latest firmware.
...
I keep getting stuck around 880-940
You'll be hard pressed to find anyone considering that to be "stuck" or subpar performance. Those numbers are within "spec" for sustained gigabit speeds, and depending on how you're measuring these speeds (speedtest browser vs. speedtest cli, time of day, speedtest servers), the results will vary greatly.
You also have to factor in the physical limitations of the switch. Local speeds (which depending on how your vlans are setup) bypass NSS altogether (it's not needed) and will not reach true "1Gb". Even locally, with jumbo frames enabled, the most I'm able to achieve is 990mpbs.
Sagemcom F3896LG directly connected i get flat 1000mbps.
I looked over the spec sheet, the Sagemcom has a 2.5Gb WAN port, while the R7800 WAN and LAN are both 1Gb. The 2.5Gb WAN port is likely able to cope with some of the TCP overhead, which would account for you seeing 1000mbps in your testing.
Yeah might just have to invest in an x86 system as router because of next years fiber operation with an ISP that can offer 10gbps.
@ACwifidude
I come to an issue with one of your latest images:
I know for a fact i have way more space that shown here, what happen and how can i fix it?
Just to notify, i have been battling with this for quite a while.
This router has a mount point that is 4GB and people been saying this is for the default firmware.
But i refuse to believe that the default firmware and software requires 3GB.
Is there any possibly way to figure out what is within this area and if i can eliminate what ever is in it to use for my own?
Another thing i found out is that wifi is using the main cpu cores on the NBG6817 firmware.
I can only reach up to 500mbit with a cpu usage of around 41%.
I can only reach up to 500mbit with a cpu usage of around 41%.
I highly advise reading over this thread, I know it’s a lot of posts, but well worth the scroll.
Currently, we’ve only been able to achieve partial NSS acceleration for wifi on kernels 5.4 and newer. Stock is able to achieve full acceleration but the trade off is being stuck with an ancient kernel (3.14) with zero community support or @quarky’s original 4.4 build which was able to achieve full NSS acceleration. However, he’s since moved onto 5.4 and the community as a whole has moved to 5.10/5.15.
I will add, 500-600mbps that’s been achieved with partial NSS acceleration is comparable to the speeds achieved on stock firmware, at the cost of CPU cycles.
QCA is also not actively developing on ipq806x/ath10k platform anymore. The attention is more on ipq807x/ath11k/ath12k, therefore doubtful we’ll see any more improvements regarding NSS acceleration. Especially since the source for the firmware is closed source (not drivers, which is open)
I just checked that i'm not on the ath10k firmware, just switched to this firmware but now i have this problem:
I can't seem to figure out what is wrong.
Tried disabling them, removing them to make new ones. Nothing works
-UPDATE-
Just got worse.
Just reflashed with newest ath10k firmware without keeping my settings, now the wireless tab is gone plus all the configs.
Tried to install kmod-ath10k-ct but says: Kernel: Version incompatible.
Tried to do it with the opkg install and github link didn't work, downloaded the file and did the opkg install /tmp/filename(kmod-ath10k-ct) and still is throwing errors:
I'm lost for words, i don't even know where to start to solve this problem.
This is (again) normal behavior for these routers. @qosmio provided a good suggestion to spend some time reading through this thread and familiarizing yourself with the standard operations and expectations for ipq806x NSS builds. It’s a wonderful system for many users, but perhaps your needs/wishes go beyond what it can provide.
Unfortunately many networking functions still are single threaded so 50% could mean one CPU is near 100%, the other is at idle. Common NAT speeds should be between 880-940 when testing against public servers. Should be upper 930’s - 940 solid if testing with iperf at home (I’ve done this to prove the issue is your ISP or somewhere else in the public sphere). Wifi speeds can be tested similarly, high 500’s (rarely low 600’s) is a typical speed for a 2x2 newer client on 80Mhz channel before you max out the CPU.
Kmods modify the kernel and create a different hash as a result. I post all my packages but unfortunately despite the original packages being the ones built for the custom kernel- frequently opkg has kernel incompatibility errors with custom builds like this one that prevent installation of individual packages. Frustrating right?
The cleanest way to have all the packages you want is to build from scratch. It seems daunting but if you have a Linux computer you can follow the instructions in post #2 above to get exactly what you want.
Damn, that's Rock stable.
No idea what mine would have run before.. Usually I do manually restarts every now and then, but 41d is for sure freaking amazing.
@ACwifidude should the dependency cycle solved? I noticed qosmio saying that it's solved and we just need to refresh the NSS feed and I saw you added that to your repository. While compiling I got an error with the NSS crypto package again though.
You should use the most recent image @ACwifidude created. About 2 days ago. It includes the commit that reverted enabling Krait CC which causes this issue.
I tested out latest build on Kernel 5.15 and got reboots after about 2 mins uptime. flashed back to kernel 5.10 and all is stable again. My r7800 is running as a dumb AP and I did save configs on flashing.