Unifi POE 24 switch, dead ethernet ports

[Krieger]

Hi all,

I just purchased a "non-working" Unifi 24 port POE network switch. Some of the ports on it work, some of the other ports are dead. They data light on the port stays a solid green. I disassembled the switch and did not see any burned traces or burned components. It appears that the POE portion of the ports work, it's just the data connection that is not working. I'm not familiar with repairing networking gear, and I was wondering if anybody could give me any pointers on what I should look at, or where I should start troubleshooting. I can post pictures (I'm assuming that will help).

From my initial teardown, it seems that there is a daughterboard for power delivery, and the main networking components are on the mainboard. I also have a non POE version of the switch is that would help in troubleshooting at all.

Thank you

[asis]

Hi,
What is missing on RJ45 connectors - 48V or LAN (Ethernet)?
Photo is needed.

[Whales]

Ethernet is missing.

I'm not certain, but I suspect 24 port gigabit switches might use a tree hierarchy of multiple switch chips?  Eg lots of 5-port or 8-port gigabit ethernet switch chips, then a layer more above those?  See if the non-working ports form a distinct pattern of (eg) 4 non-working ports.

EDIT: Also check the switch configuration if it's a managed switch!  Those ports might be software disabled or otherwise forced into unusable link modes.

[Monkeh]

I'm not certain, but I suspect 24 port gigabit switches might use a tree hierarchy of multiple switch chips?  Eg lots of 5-port or 8-port gigabit ethernet switch chips, then a layer more above those?  See if the non-working ports form a distinct pattern of (eg) 4 non-working ports.

You'd think, but no, that would introduce arbitrary bandwidth restrictions between port groups. It will have one colossal SoC with pretty much everything but 10Gb uplinks integrated. If there's a failure past the magnetics, it's not repairable - unless of course you have a line on replacement SoCs and the tooling to successfully replace a ~1000 ball BGA.

I would verify the integrity of each transformer and if they're good, scrap it or forget the ports exist. After, of course, verifying the switch isn't misconfigured.

[Krieger]

Ethernet is missing.

I'm not certain, but I suspect 24 port gigabit switches might use a tree hierarchy of multiple switch chips?  Eg lots of 5-port or 8-port gigabit ethernet switch chips, then a layer more above those?  See if the non-working ports form a distinct pattern of (eg) 4 non-working ports.

EDIT: Also check the switch configuration if it's a managed switch!  Those ports might be software disabled or otherwise forced into unusable link modes.

So, I'm not sure what the "switching" chips are that you speak of. Here is a video of a teardown of a similar switch (I have the POE version of this switch, but they are almost identical). https://youtu.be/bnyCDTA6sME?t=221. I was thinking that it could possibly be the g48209sng chip that is bad. The switch is a managed switch, it was factory reset. Currently, there are 6 dead ports, possibly more that are on their way out. There is a block of 4 ports, and then 2 ports that are not near each other. Seeing as there are 12 chips, it would make sense for blocks of 2 ports to go bad.

The ports with red plugs are dead. They get POE, no data connection.

For what it's worth, when I was testing the ports, some other ports acted odd, only getting 100FDX instead of the full gigabit connection. I went back and tested them again, and now some of those ports that were acting oddly are now working at 1000mbps, so it could be a SOC issue. I'm not sure.

[Monkeh]

Those are not chips, they're transformers. The only chip involved is the gigantic one under a heatsink.

[Krieger]

Any idea how I can check to see if the transformers are bad or not?

[pcprogrammer]

For what it's worth, when I was testing the ports, some other ports acted odd, only getting 100FDX instead of the full gigabit connection. I went back and tested them again, and now some of those ports that were acting oddly are now working at 1000mbps, so it could be a SOC issue. I'm not sure.

The difference between 100Mbit and 1Gbit connections can be due to the cables. 100Mbit works over just two pairs, while 1Gbit needs all 4 pairs connected.

For testing the transformers I guess you could check signals with a scope, but you need at least a 100MHz bandwidth. The frequency on the pairs is only 25MHz.

Do you have some insight on the history of the switch? It might have been damaged by thunder and is slowly degrading. I had that happen once. Directly after the strike only one port failed to work, but a while after more started to fail.

[Monkeh]

Easiest way to test the transformers is continuity across the windings. There's not much to go wrong short of them going open circuit.

[Krieger]

For what it's worth, when I was testing the ports, some other ports acted odd, only getting 100FDX instead of the full gigabit connection. I went back and tested them again, and now some of those ports that were acting oddly are now working at 1000mbps, so it could be a SOC issue. I'm not sure.

The difference between 100Mbit and 1Gbit connections can be due to the cables. 100Mbit works over just two pairs, while 1Gbit needs all 4 pairs connected.

For testing the transformers I guess you could check signals with a scope, but you need at least a 100MHz bandwidth. The frequency on the pairs is only 25MHz.

Do you have some insight on the history of the switch? It might have been damaged by thunder and is slowly degrading. I had that happen once. Directly after the strike only one port failed to work, but a while after more started to fail.

I have no idea on the history of the switch. I would assume lightning strike. If I end up with a cascading failure then oh well. I basically got 70% off the going rate, so if it lasts me a few months to a year then I'll be happy.

[NiHaoMike]

Have you checked the caps?

[Whales]

Don't discount dirty RJ contacts!  Get a toothbrush in there, the gold plating may have rotted (in which case cleaning them is only a temporary fix).

By ethernet switch ICs I mean things like Atheros AR8327 (datasheets).  Combined with a power supply, a PCB, the transformers ("magnetics") and the RJ-45 sockets you can make a fully functional (dumb) network switch.

[TheMG]

I can't seem to find pictures of this switch with the POE board removed, and am not familiar with that particular switch.

Some switches use ethernet PHY chips separate from the main switch, in which case you might be lucky and replace the chips associated with the troublesome ports. Some switches everything is integrated into the main switch chip in which case you're probably out of luck.

But I do agree with the assesment that the switch was likely the victim of lightning damage. It's possible, though unlikely, that the magnetics got damaged. I've repaired a few lightning/surge damaged ethernet devices and not once have I found the magnetics faulty despite the ethernet interface getting fried.

[Krieger]

I can't seem to find pictures of this switch with the POE board removed, and am not familiar with that particular switch.

Some switches use ethernet PHY chips separate from the main switch, in which case you might be lucky and replace the chips associated with the troublesome ports. Some switches everything is integrated into the main switch chip in which case you're probably out of luck.

But I do agree with the assesment that the switch was likely the victim of lightning damage. It's possible, though unlikely, that the magnetics got damaged. I've repaired a few lightning/surge damaged ethernet devices and not once have I found the magnetics faulty despite the ethernet interface getting fried.

The switch looks pretty much identical underneath the POE board. I don't think this has ethernet PHY switches on it, unfortunately. Would damaged magnetics cause the issues that I am having?

[thm_w]

From the exterior RJ45 socket, with the tab facing down, measure the resistance between the leftmost pin to the other pins.
Compare this between one bad port and one good port, see if you notice any major differences.

I've seen mostly IC failures but maybe 20% where the magnetics have failed, but, this is on non-POE port.

[AndyBeez]

Have you updated (reflashed) to the latest firmware? Unify [Ubiquiti] switches are 'cloud managed', so I am wondering if you have a ghost config tapped in the NAND? It would be logical to have six ports locked-out/unconfigured or, they might be for some other weird VOIP use?
Not knowing the model number, cannot RTFM
Otherwise accept it's half a brick and spend $30 on another switch.

[Krieger]

Have you updated (reflashed) to the latest firmware? Unify [Ubiquiti] switches are 'cloud managed', so I am wondering if you have a ghost config tapped in the NAND? It would be logical to have six ports locked-out/unconfigured or, they might be for some other weird VOIP use?
Not knowing the model number, cannot RTFM
Otherwise accept it's half a brick and spend $30 on another switch.

Yea, config is wiped and firmware has been updated. Can't exactly buy one for $30, they go for $400 new.

[AndyBeez]

$30 bucks maybe a bit low ball. The problem with 'cloud managed' network hardware is although you can buy it cheap (the fantastic Cisco Meraki MS220 switch is on ebay for around $120) the licensing will have expired. Which means servicable devices are just e-waste. Other than either buying a licence, or reflashing with OpenWRT, these bricked by firmware boxes are a tragic waste of good parts. All because manufacturers don't care as they are in business to sell boxes with no respect for the planet. I've had stable performance from Netgear PoE managed switches. Just a web interface and no cloudy BS.