Tuesday, 27 May 2014

Unifi UAP-AC with 802.3af and 802.3at PoE switches

This is just a warning to anyone using the Unifi UAP-AC Access Points and not using the supplied PoE injectors.

These are really good APs, and we recommend and sell them. However, I was caught out by a slight problem this last week.

We are running the latest beta test code for these. This is, of course, a risk we are taking. We fully understand that risk. But what happened was not actually an issue with the beta code.

We upgraded to 3.1.12 from 3.1.10 and encountered a problem. The APs kept crashing. We were lucky if they ran for an hour, and often way less. This made the WiFi almost unusable.

It had been fine on 3.1.10, or so it seemed, so we assumed it was an issue with the new beta s/w. Shit happens, we are running beta s/w, so not a surprise. Ubiquiti tried hard to find the issue. We sent logs and config and so on. It was clear from the beta forum that others had the same issue. The PRO models were fine, only the UAP-AC units were crashing.

We found the problem, to our surprise.

These come with there own in-line power over ethernet (PoE) power supplies. Some of the lower models are a passive PoE, but the PRO and AC units claim to be 802.3af and 802.3at compliant. To be honest we had not done a lot of PoE, and we got some nice PoE switches. What we had not spotted is that they are 802.3af only, and that is only 15W of power. The UAP-AC units need 22W so the 802.3at (25W) switches are needed.

Now, my understanding is that 802.3af and 802.3at should negotiate the power requirements. This should mean that a device needing more than 15W should not negotiate with an 803.2af switch and so should not work at all, perhaps indicating a power fault. Sadly, either the switch or the AP does not do this and it tries to work anyway. Somehow the UAP-ACs were working but the s/w update clearly increased the power usage and caused crashes.

Having worked this out, changing to use the provided in-line supplies fixes the issue. This is messy when we have PoE switches anyway, so we are going to try and find some 802.3at PoE switches.

It is a shame. If the negotiation logic had worked we would have realised our error when installing these, rather than much later on a s/w upgrade. As I say, I don't know if the switch (netgear) or the AP is to blame (though latest comments on the forum suggest it is the AP), but it is a cautionary tale none the less.

7 comments:

  1. Most likely the AP I Would have thought.

    ReplyDelete
  2. Almost no equipment gets power negotiation right, most don't bother with it at all. At work we make devices which have USB connections and can supply no more than 500mA, you're supposed to negotiate if you want more. But no equipment does. We're exploring the options, but may have to resort to hardware that simply shuts down the power a little above 500mA. This is a battery powered device, we can't have other kit pulling arbitrary amounts of power.

    ReplyDelete
  3. My understanding is that it works by the switch supplying power at around 20V with a 100mA current limit and then measuring the current consumption and working out the resistance. This resistance is used to indicate the power rating.

    Its optional however and I suspect if the current draw is outside of the range for 802.3af then the switch assumes class 0 (rather than only assuming class 0 is the power consumption is practically zero). That would mean that a 802.3af switch would always try and supply power anyway to a device requiring more than 15W and cutting off power if it does take too much.
    Either way the AP might not be supporting power negotiation or if it does then it could be the actual standard which is causing it to partially work on 802.3af switches.

    ReplyDelete
  4. We had similar problems with Zyxel APs on Netgear POE switches. We use rack-mount in-line injectors now instead.

    ReplyDelete
  5. Cisco APs have a similar issue, but behave correctly - if the power is not enough, they will disable some or all of the radios until they are connected to a better supply. Can just sometimes be hard to spot why your new AP is up and running but not actually providing service, but that's our monitoring, not the system's fault.

    ReplyDelete
    Replies
    1. That sounds familiar, I recall reading of another PoE AP which had something like 3 radios, but could only power two of those from 802.3af. If the Unifi units rely on firmware powering something down to be inside the 15W limit, but the beta firmware fails to do that, it would explain the whole problem.

      Have you reported it to Unifi to see what they say? (For that matter, is there any other difference now you've switched to .3at, like better range or something that might fit with something having been enabled that wasn't working previously?)

      Delete
  6. Thanks for the heads up. I have a number of AP-AC units in the field. Some network using the In-Line PoE+ injector, and some not. I have not had any problems with these units(as a some sites are fine without the PoE, until last night, it just kept crashing. I thought it was a PoE issue, thanks for your clarification.

    ReplyDelete