I found one difference in how the two devices are shown on pci bus
cat /proc/bus/pci/01/00.0 |hexdump
on working box:
0000000 8086 09a9 0546 0010 0001 0280 0010 0000
on not working box:
0000000 8086 09a9 0546 1010 0001 0280 0010 0000
The value is the same if line is up or down.
Are the higher 4 bit in the value at byte position 6 related to tc_status as defined in vrx518_tc_drv-ugw_8.5.2.10/grx500/inc/tc_main.h?
Is this a read only value? Or could the device be forced to run state in setting it to 0?
Maybe the failed clk is only running in case device is in run state...
Thanks for the work.
While my attempt to build it myself failed, yours works nice on a FB7520.
I'd like to use htop, but when i install it i get various error messages like "Multiple packages (libgcc1 and libgcc1) providing same name marked HOLD or PREFER. Using latest." and more.
But htop is installed. When i try to start it, it says htop: not found.
Well, it's there (/usr/bin/htop). I don't know if it's an incompatibility with some libraries or whatever.
Do you've got an idea how to make it work?
Build the image yourself with htop included. This can happen on snapshot builds if you don't integrate packages into the image. The older the build is, the more likely it is to happen. The libraries in the package feed changes while the ones on your device stay.
I updated the build to contain htop
same link as above to get it
I'm not sure what to tell you to help you build it yourself, I basically merge hauke/glibc , dhewg/vr11 and ldir/mine there's other patches i'm using but basically that's it.
I flashed the not working box with freetz-ng ( standard AVM image + openssh server).
The AVM kernel is performing a different init of the vrx518.
hexdump of proc/bus/pci/01/00.0 shows some values different after avm init:
Thank you very much.
Now i can see that a download at 100MBit/s makes approx 25% processor load (2 cores with both approx. 20-30%).
That's clearly a big advantage compared to the HH5A.
Now there're enough resources left to think about SQM/CAKE and other things.
Well, i tried this but the router didn't like my build.
I'm quite sure i might have made one or two little mistakes...
When i have some more spare time left (this might take a while) i'll start from scratch.
Until then i'm quite happy with the build made by @wilsonyan.
I'm very grateful what you and all the other delevopers do and achieved by now.
it might be better than that with software offload turned on (optionally add "nft add flowtable inet fw4 ft { counter }" to your /etc/rc.local or system -> startup -> local startup in the web interface for traffic counting)
also i'm not 100% on whether the irq affinity script I put in spreads the irqs out completely optimally
I can see there's a difference, but the load on the 4 cores changes so quick that i'm not sure if there's really an improvement or just my imagination.
Maybe if the router would've more to do than "just" PPPoE and routing/firewalling, the effect would be more significant.
I'm just happy that there're now enough resources left for maybe a speed upgrade and/or traffic shaping and other things that might come.
I just think it's quite strange that here in Germany you can get a used Fritzbox7520 for less money than a pure (S)VDSL2-modem like a Draytek Vigor. More bang for less bucks .
That's where i got the 7520 from...it's the 1&1 branded version.
No long time measurements, just spot values:
With DSL and 2 LAN ports connected, WLAN disabled: 5,6W.
With DSL and 2 LAN ports connected, only 2,4G WLAN enabled, 2 clients: 5,8W.
With DSL and 2 LAN ports connected, only 5G WLAN enabled, 2 clients: 6,0W.
With DSL and 2 LAN ports connected, both WLAN enabled, 2 clients: 6,2W.
Measured with an ELV Energy Master Expert.
Thanks, these numbers look pretty close to my HH5A operating as bridged modem (actually your numbers are a bit lower, but only a bit), so probably time to see whether I can get a FB7520 on the cheap to profit from the better sync and higher CPU performance (with my 116.7/37.0 sync I get the expected throughput unless I try to bi-directionally saturate the link in which case the aggregate throughput looses something like 10% over the theoretical upper limits (already accounting for ACK traffic) nothing that gives me much concern but I am curious whether fast CPU will fix this).
That's right.
But i consider such high frequencies on our (very) old telephone lines as ridiculous.
So it's good to see that nowadays many lines are upgraded from copper to fiber. But of course not everywhere...where i live, VDSL is available only since end of 2019.
Maybe i'll witness that i'll get fiber into my flat within the next 10-15 years...
Even worse in Austria: my VDSL2 line is maxed out at 40Mbit/s with no official plans on getting fiber ("There is 5G coverage, why do you need fiber?").
My 7520 is also an imported 1&1 version, they are ISP-only models. Our providers use branded 7530 boxes (with fake or flaky NAND ICs, at least the one A1 version I had for a short time).
Yeah, some people think that 5G is the solution for everything.
Someone should tell them what's possible with 5G and what is not.
That should be an exception...they're still being produced by AVM, and i don't think they would risk their reputation by (knowingly) using unreliable or even defective chips.
With a bit of luck you can get a used 7520 for less than 50€ (incl. p+p).
You just have to bear with the black case and the slightly outdated design of it. I do like the look of the new case (like the 7530 and 7590) better, but not at the cost of additional 20+ €.