Construction fails in networking gear (#R9)
Ah well, I’m too lazy to craft another post dealing with things currently in development – let’s rant instead.
A few weeks ago I bought this huge lot of “failed” networking electronics that finally led to having an 8-port PoE switch at home, which is a nice addition – the MikroTik wAP ac finally runs on PoE as intended, the four ports (2x1G, 2x100M) of the Fritz Box router are no longer overcrowded, I did have a good time repairing two of those switches easily and I didn’t even spend a penny on all that thanks to a few sellable items (plus brownie points for donating quite a few more).
Naturally not all gear is great. There’s turds everywhere, and this lot had everything from noname China to Cisco.
The Netgear R6120 “AC1200” WiFi router
This was a severe case of Murphy’s law. The router didn’t power on, and I finally trashed it for having a bad CPU. How do you open them? Easy, look for screws on the bottom. There aren’t any visible? Check underneath the rubber feet. Still none? Well, maybe just pry it open carefully, might have clips all around, dunno…
Yeah, you tell me which TWO rubber feet I did remove first and then had a go at breaking it open, as “there are no screws hidden below”. Fucking Murphy.
But it gets better. Meet:
The D-Link DWR-116
I can’t remember what was wrong with that one, but it got trashed as well. Same procedure, but I found…three screw holes?
Oddly enough, there’s no reason to not have the fourth one in place. The bottom PCB layer is empty, and the top one has an oversized power trace in the vicinity. Don’t think there’s something important in the middle layers.
Did the layout guy just fuck this up and then the case designer made this mounting post optional? I’m not sure…
China access points
So this is two extremely cheap APs with two quirks. First, please do compare the cases yourself.
PCB’s different (entirely different chipset actually, Mediatek on the WAVLINK WS-WN518N2-C-V1.1 and Realtek on the even more noname “NetVIP” product), different antenna situation as well. But the case is identical except for the color/print. That’s really the bottom end of Wan Hung Lo manufacturing – designing an entirely new product that fits an existing case mould and has the same specs. Geez.
Quirk two: Both of them failed because the universal power plug situation literally fell apart. This is the mains contact:
And this is the case where the local power plug will be fitted into:
Nope, not going to be used anywhere I can grab this piece of shit and throw it away. Both of them cracked and failed before actually going up in flames or something. The AP PCBs got donated with some other stuff and the guy actually had a deeper look into it (they use basic 5V supplies, up to about 300mA under load, so that’s handy). Turns out they have open telnet access on port 2323 and they ping A LOT. They also start up with their own IP network and then remember their settings, stopping every (possibly conflicting) service and using the user settings instead.
And the last one for today:
A Qivicon Home Base
This is some Smart Home base station gadgety thing that is (or was) apparently sold by Deutsche Telekom, one of the major German ISPs/telephone companies. 3rd party crap for some buzzword bingo. Once activated it is tied to your account, reselling seems next to impossible, and it doesn’t frigging work without you not only paying for the device (30 bucks?), but also forever on a monthly basis. If that’s not enough of a red flag, here’s the power jack:
Yeah, of course it needs a custom power jack, because only mega expensive original power adapters are good enough to power this turd.
Any reason for this? Nope. Not a rare voltage (12V – they need to improve that! Maybe 11.2V AC?), not a high current device (2A I think, mainly for all the USBs), no need for a coding pin as there isn’t a different version with less/more inputs and stuff. It’s purely there to annoy the customer that accidentally wrecks the supply.
Tossed. Well, ripped apart and tossed. I tried desoldering the inductors (WHAT DOES THIS THING NEED FIVE VOLTAGE RAILS FOR?!) and there has to be more than one major ground plane beneath. Soldering without additional heating is next to impossible with an 80W iron. This might even be a 6 layer board where everything is via-stitched together for heat dissipation or EMC issues. Crazy.
On the contrary, repairing a lot of upscale Netgear products was a breeze, a lot of the small APs use interchangeable power supplies, power plugs, even PCBs/firmware (I did a brain AFK oopsie once and it worked). This is one of the PoE switches where one of the SMD capacitors on the internal power supply failed (shorted), causing this to never power on. The other one had a dead SMD diode nearby, causing the same failure mode. This was both on the 54V PoE rail, so getting replacement parts isn’t as easy as grabbing the next best component in the parts drawer, but nowhere near impossible.
Shiny leaded solder and a bit of transformer touchy-touchy with the soldering tip later, both of them came back on, each tested 1TB+ of iperf3 data over each port combination, perfect!
Mine’s now running with a MeanWell 48V 1.25A supply instead of the 54V Netgear ones, which is both fine for PoE (42.5 to 57V for 802.3at) and the switch itself (needs something above 35V to even turn on). I also interchanged the pitted power jack on mine with a standard 5.5/2.5mm one, as Netgear uses rare ~6/3mm barrel connectors. To their defence the top level power supply has 135W, so at 54V nominal that’s 2.5A. That’s not stretching the amperage limits of those barrel connectors to unrealistic levels, but it’s a significant load for 24/7 operation at elevated temperatures, and Jon Doe connecting 54V to his 12V doodad might be a bit more catastrophic than other common voltages with the same matching connector. For my max 1.25A that’ll do just fine, and I didn’t need to modify the power supply. Plus I’m the solely responsible idiot that needs to fix my very own gear if I accidentally kill it with 48 volts…
[…] third working PoE switch after minor repairs, but also added quite a few dead PCBs to the tree. See Construction fails in networking gear (#R9) for […]