Mitigating Pause Frames

[mhoppes]

Throwing this out in the public forum for discussion. I *believe* this is how Chris has his network setup, but don't quote me:

To use a Netonix switch as an inline PoE injector for backhauls and a PoE injector for APs, but to have a router at each site a setup similar to this could be used:

Bring an AF24 into Port 1.
Plug Router eth1 into Netonix Port 23 (inbound)
Plug Router eth2 into Netonix Port 24 (outbound)
Plus APs into ports 5,6,7 in Netonix.

In theory, I should be able to tag outbound traffic on Port 24 (vlan 5), and then untag on ports 5,6,7.

For inbound though, I should be able to bring the AF24 into the switch, then tag the traffic into Port 23, correct?

I'm assuming with this setup Pause frames should be mitigated because the traffic is hitting the router, even though it's still going through the switch, before it goes to the APs?

Is that a correct thought process?

Similarly, I would think you could still bridge your management traffic direct from the AF24 to switches and other devices, since that traffic would be extremely small compared to your data traffic.

The biggest question is, even though the traffic flow is:
AF24--->Netonix--->Router--->back to Netonix--->AP
and it flows through the Netonix, the pause frames should be mitigated because the router will act as the TCP buffer device... yes or no?

[tossman101]

Just to clarify how Netlinx is set up:

At each main tower there is a Cisco router and a 24 port Netonix switch. The switch is a mid-line power injector as you said, the back-hauls come into a high voltage port (ie. ports 1-4), are bonded to another port which then plugs into the Cisco. So an AirFiber plugs into port 1 gets bonded to port 25 (a copper SFP) which then plugs into a routed port on the Cisco, the bonding is done by placing the two ports in the same vlan where both ports are untagged. The APs come into the normal voltage ports, these ports are each a separate vlan (untagged) for each AP. Ports 23 and 24 are setup as a static lag and plug into routed ports on the Cisco, these lag ports are tagged in each AP's vlan to carry their traffic to the router.

The setup you're proposing, is that with just one back-haul? If so I believe your theory of how to tag/untag the vlans should work and I would imagine help in mitigating pause frames. We do not use a separate management vlan for our back-hauls/APs/CPEs so I cannot speak to how that would work.

[mhoppes]

I spoke with Chris and basically what's going on here (as I suspected but wanted to confirm) is the router acts as a buffer/d-marc point for the pause frames. Essentially any pause frame generated by the APs or the switch stop at the router.

By using a different switch port and router port for each additional backhaul coming off of a site you also help break up the pause frames so you don't have one huge breathing network.

[sirhc]

CLICK IMAGE BELOW TO VIEW FULL SIZE

Tower Setup.png (82.22 KiB) Viewed 21179 times

This way traffic going into the tower destined to the next tower will not be affected but the pause frames generated by the "local" radios on that tower serving as AP's or private PTP links to higher end customers.

Also since all the "local" radios on the tower go through a LAG to the router there is a 50/50 chance that the pause frame issued because one AP's port buffers are full will even affect other local radios/streams on that tower as the switch will issue the pause frame to only one of the interfaces in the LAG, the one that is feeding that AP at that time.

I go over this in the WISP Switch movie on You Tube: https://www.youtube.com/watch?v=8JvBEAD4MFM

[sirhc]

We break up each port into a separate VLAN which allows us to segregate the IP's of each VLAN on the router to a separate virtual interface which isolates each AP from each other.

We want as small of a Layer 2 bridge as possible. This way broadcasts and other type of possibly bad traffic from one AP never sees another AP.

You can see our VLAN Tab in the movie in the Youtube movie mentioned above.

[sirhc]

Will you see pause frames on the interfaces connecting to the airFIBER units?
Yes, you will see "a few" Rx Pause frames as the switch interface is 1G but the AF transmits at less than 1G so there are moments when the switch will try and send too many packets at the AF and the AF will ask the switch to pause. You will however never see Tx pause frames on that port unless something is wrong as the AF will never be able to send 1G of packets at the interface and thus its buffers will never fill up (unless something is wrong).

[mhoppes]

Screen Shot 2015-10-08 at 1.09.03 PM.png (23.77 KiB) Viewed 21134 times

Hello. My name is Matt. I'm a Pause Frame Addict and I have a problem. *sits back down*

[sirhc]

I am guessing you have 1 or more weak links with low CCQ or AMQ and that customer will cause the AP to hang on that customer. Also if the AP is busy on 1 bad link and other clients request data this can cause a flood of packets to hit the interface.

Another thing can be a customer using Torrents which can sends thousands of packets at the interface from thousands of different sources and can also overwhelm the port buffers forcing it to send a pause frame.

Or this can be a symptom of a flat network, or all of the above.

Really am excited for you to move to a routed network topography, you keep hinting about it being in the works.

[mhoppes]

And a head end traffic shaper....if I can get this MikroTik to work for more than a few days at a time.... Piece of junk.

[mhoppes]

I also just learned Netflix will often attempt to overwhelm an end users connection by sending multiple steams of data above the data rate the CPE can handle.

Sigh.