Thanks for the clarification.  That sounds like VoIP should be happy.  People 
tend to obsess over latency when it comes to VoIP, but in my experience packet 
loss is what will cause problems.

What I was thinking about was if there was no ARQ mechanism, I should set the 
max modulation as low as possible to still get the required throughput.  But it 
sounds like this should be unnecessary.

It does sound like we need to figure out how to set 802.1p as well as DSCP, and 
I think this may have to be done on a hop-by-hop basis, I’m not sure Mikrotik 
propagates ingress priority to egress priority.  Currently we are using a 
mangle rule and address lists to set DSCP on VoIP packets where they enter our 
network.


From: Chuck Macenski 
Sent: Saturday, September 10, 2016 2:23 PM
To: [email protected] 
Subject: Re: [AFMUG] airFiber error correction

That should read does not do DSCP.... Darn keyboard errors...

On Sat, Sep 10, 2016 at 2:22 PM, Chuck Macenski <[email protected]> wrote:

  AirFiber prioritizes traffic based on the 802.1 VLAN tag priority bits. Each 
of the priorities is sorted into unique MAC queues with the highest priority 
queues supplying data first to the over-the-air multiplexed data stream. That 
stream is then sent using a selective repeat ARQ mechanism - the RF symbols 
carrying this traffic are encoded using FEC coding. AirFiber does look at the 
IP header and thus does not do DHCP.

  On the decoding side, the multiplexed data stream will not skip a missing 
fragment - no packets from a given VLAN priority are ever delivered 
out-of-order. The mean latency adder for fragment re-transmission is roughly 1 
- 1.5ms for a 2ms frame. This number varies based on the radio type, configured 
frame length, and whether the link is half or full duplex.

  Chuck

  On Sat, Sep 10, 2016 at 10:20 AM, Ken Hohhof <[email protected]> wrote:

    Oops, I mistakenly thought that was an update of this article:
    
https://help.ubnt.com/hc/en-us/articles/205231970-airFiber-How-does-airFiber-handle-QoS-and-frame-prioritization-

    If I take that at face value, airFiber uses 802.1p but not DSCP.


    From: Mike Hammett 
    Sent: Saturday, September 10, 2016 10:00 AM
    To: [email protected] 
    Subject: Re: [AFMUG] airFiber error correction

    airMax != airFiber




    -----
    Mike Hammett
    Intelligent Computing Solutions

    Midwest Internet Exchange

    The Brothers WISP






----------------------------------------------------------------------------

    From: "Ken Hohhof" <[email protected]>
    To: [email protected]
    Sent: Saturday, September 10, 2016 9:50:50 AM
    Subject: Re: [AFMUG] airFiber error correction


    According to the chart here, airFiber does prioritize based on DSCP tags 
(which are visible at layer 1):
    
https://help.ubnt.com/hc/en-us/articles/205231750-airMAX-How-is-QoS-and-prioritization-handled-by-airMAX-

    Typically VoIP equipment would set DSCP on outgoing packets, and border 
routers would set DSCP on packets inbound to our network based on criteria like 
src or dst IP address.  Then switches and bridges can prioritize based on those 
tags.

    airFiber doesn’t appear to be as sophisticated as some other products, in 
that you can’t configure the priorities, and the stats (at least in the GUI) 
are kind of meager.

    I’m mainly trying to determine if airFiber has an ARQ mechanism, the spec 
sheet doesn’t seem to mention error correction at all.  I’m assuming the OFDM 
implementation has some form of forward error correction to monitor the error 
rate and adjust modulation and also for the coding gain.  Whether it has FEC 
block coding or ARQ, I don’t know.  Licensed backhauls typically don’t use ARQ 
because SNR is determined more by background noise than interference, plus ARQ 
requires buffering data until errored packets can be retried (assuming you 
still want to deliver packets in order), which impacts latency.  I just don’t 
know if the airFiber designers applied the same thinking, given that it 
operates in unlicensed spectrum.


    From: Josh Reynolds 
    Sent: Saturday, September 10, 2016 9:09 AM
    To: [email protected] 
    Subject: Re: [AFMUG] airFiber error correction

    Fixed delay jitter buffers are amazing things.

    That said, airfibers have NO IDEA what data they are sending. The 
management interface is aware of packets destined for it, and that's it.


    On Sep 10, 2016 9:07 AM, "Ken Hohhof" <[email protected]> wrote:

      VoIP is UDP.  Also latency variation or out-of-order packets are bad so 
there is no retry mechanism at transport or application layer.  So the packet 
gets one chance for delivery, otherwise it gets dropped, and there goes 10 ms 
of voice.

      airFiber already prioritizes packets based on COS and DSCP values, so 
VoIP packets would typically be prioritized for example if DSCP=46.  Plus it’s 
fairly common to send beacons and retries at a lower modulation.  So all the 
pieces of the puzzle would seem to already be there.


      From: Josh Reynolds 
      Sent: Saturday, September 10, 2016 4:05 AM
      To: [email protected] 
      Subject: Re: [AFMUG] airFiber error correction

      For you first part, I don't know.

      For your second part, I have never heard of a feature like this - 
normally that would be handled by TCP for most use cases and not a layer1 
device.


      On Sep 9, 2016 11:35 PM, "Ken Hohhof" <[email protected]> wrote:

        I am going to assume airFiber has some sort of FEC, but does anyone 
know if it has a retry mechanism (ARQ)?  And if it does have ARQ, what is the 
impact on latency, and can retries result in out-of-order packets?

        And has anyone ever heard of a manufacturer offering the option of 
having high priority latency sensitive traffic sent at a reduced modulation 
rate to insure it gets there the first time? 




Reply via email to