As we have performed several test with an oscilloscope attached to the APU I hereby report out results:
setup-A: 1x APU 2c4 with two Mikrotic mPCIe WiFi cards (1x R11e-2HPnD & 1x R11e-5HnD).
-configuring two APs, one on 2,4Ghz and the other on 5GHz we can trigger the power off issue in seconds by having a client connected to each AP and just performing a cat /de/zero download via ssh on each client
-the oscilloscope is attached on the 12V line via an 1Ohm measurement shunt
-the voltage monitoring across this resistor are similar to the current flow (1V equals 1A)
-the picture shows the last microseconds before the board shuts down, the overall power consumption is the delta between the two voltage lines y1 & y2, which is about 2V so respective 2A… meaning the whole APU toke 12V * 2A = 24 Watt while the two downloads are running
setup-B: 1x APU 2c4 with two Mikrotic mPCIe WiFi cards (1x Compex WLE200NX & 1x R11e-5HnD)
- same AP & client setup as before with cat /dev/zero downloads via ssh
-no power shutdown from APU and 1,6A load on the 12V line
-in addition we added an 120GB SSSD to the mSATA port as another power consumer and loaded it with some dd fileoperations while downloading
-the overall power consumption increased to 2A as shown in the next plot, but the APU did not power of and kept running finr
Conclusion so far:
-it seems that the Mikrotik R11e-2HPnD WiFi card does overload one of the internal 1,8V, 3,3V or 5V rails and triggers a power shutdown of the board
-a COMPEX WLE200NX in combination with the Mikrotik R11e-5HnD and an mSATA works at the same overall power consumption just fine
-also the less power hungry card Mikrotik R11e-2HnD Wifi in combination with the Mikrotik R11e-5HnD and an mSATA works fine
Greetings from Berlin
ps: … we do now rebuild our APU setup to replace the high power Mikrotik R11e-2HPnD cards with R11e-2HnD once.
very interesting test! I follow this thread quite a while since I setup some 7 accesspoints with exactly the same hardware as you did (except with simple SSD /no SSSD - is that the important difference?).
I did not yet get any instant power off, even under full load.
Out of curiosity I now connected my lab- power supply metering voltage, ampere and watts.
The APU 2C4 with no load but both R11e-2HPnD + R11e-5HnD active draws about: 7.3W
With R11e-2HPnD under full load: 10.2W (transfer rate about 10MB/s)
With R11e-5HnD under full load: 8.8W (transfer rate about 22MB/s)
Both R11e-2HPnD and R11e-5HnD under full load: 10.7W (! but transfer rate of 2 and 5Ghz combined: only about 13 MB/s)
and still no crashes/ power offs. Both WIFI cards do have massive mutual impact on transfer rate (due to bus contention?!) if both under full load at the same time.
my test scenario:
between client1 and server (accesspoint)
on server: nc -l -p 9000 > /dev/zero
on client1: nc server-ip 9000 < /dev/zero
between client2 and server (accesspoint)
on server: nc -l -p 9001 > /dev/zero
on client2: nc server-ip 9001 < /dev/zero
- transfer rate figures above on 2Ghz generally are quite bad because my test client (old laptop) doesn't allow for better performance. The other test client easily exceeds 19MB/s on 2 Ghz.
- the power draw heavily depends on WI-FI achieved transfer rate
- I'm wondering why I'm so far away from your maximum power draw. Is my bad combined maximum transfer rate (for 2 + 5 Ghz under full load) the delimiting factor? I could also measure voltage with a digital oscilloscope. May this would shed some light on this mystery.
- Don't you think a shunt of 1 Ohm is a little too much? The voltage drop is quite high and may enforce unwanted shutdowns.
I now swapped both client laptops (i.e. the WI-FI bands). This out of the blue increases the combined (2 + 5) transfer rate to some 29MB/s(!). APU Power draw then reaches 12W. It appears the test heavily depends on the WI-FI clients . When I get some time I will repeat the test with 3 APU 2C4. 2 configured as clients , one as accesspoint.
as promised I now modified my setup to boost WI-FI performance a little:-) The 2 laptops (running before as 2Ghz/ 5Ghz clients respectively) were replaced by another APU 2C4 each. Resulting in:
server: APU 2C4 with R11e-2HPnD + 1x R11e-5HnD
client1: APU 2C4 with R11e-2HPnD (2Ghz)
client2: APU 2C4 with R11e-5HnD (5Ghz)
running no client (server idles, but both bands active): 7.5W (power draw server)
running client1 (2Ghz) gives: 22.9MB/s at 10.5W (power draw server)
running client2 (5Ghz) gives: 25.1MB/s at 10.3W (power draw server)
running both clients gives: 17.6MB/s (5Ghz) and 19.8MB/s (2Ghz) at 12.9W (power draw server, all values fluctuate)
still no crashes. Performance on both bands 2Ghz and 5Ghz deteriorates somewhat when used massively in parallel. Probably since some bottlenecks on the miniPCIe bus? But IMHO performance figures of this little box are quite well. Even under heavy work load.
Out of our 25 APU 2c4 node, there are 10 boards that face the power off isue when full WiFi download happend and if equiped with two Mikrotic mPCIe WiFi cards (1x R11e-2HPnD & 1x R11e-5HnD). So it seems some boards can handle this specific power draw and some not.
For the overall power consumption the max effective power draw at full downloads from 2,4GHz and 5GHz show 1,64A*12V=19,68W
And when we have exchanges Mikrotic R11e-2HPnD with a COMPEX 200 and so some file transfer load while full 2,4 & 5GHz download the max. effective power draw show: 1.85A*12V=22,2W. But as already said, the board does not crash in this case, even that the overall effective power draw is higher. So the two Mikrotic R11e-2HPnD seems to consume too much power an some of the sub voltages on the mPCIe interface causing the circuits to switch off on some APU 2c4.