I wouldn't do all the shaping at the AP.. I can't see that scaling well. For best performance, you'll need to do it both upstream (provider edge) and downstream (customer edge, ala on the CPE/SM).
This can present a management burden if you haven't developed or acquired the tools to do that. On Fri, Nov 25, 2016 at 12:58 PM, Ken Hohhof <[email protected]> wrote: > What I’m really trying to determine is how Cambium handles the rate limited > traffic, not the high priority traffic. > > > > The CDN and Windows Update traffic issues just keep getting worse. We > currently don’t use the radios for rate limiting, we do traffic policing > (RED) in the tower router, either Cisco or Mikrotik. > > > > Trying to figure out if there’s anything we can do short of > Procera/Sandvine/Saisei. Our network topology doesn’t have a nice NOC with > a controlled environment with all bandwidth flowing through it. > > > > I was thinking if we also implemented QoS at the AP, if it smoothed out the > bursts and put some back pressure on the CDNs in the form of increased > roundtrip latency, it might help. > > > > Mainly interested since they added max burst rate. I think as we have > deployed more 450 sectors, it has bitten us in the ass, because a 3M or 6M > customer can burst at 50M for a few seconds. Plus the router traffic > policing propagates bursts while discarding some packets, and the CDNs and > video streams seem to look at that as an invitation to just push more > traffic and retry whatever doesn’t get ACK’d. > > > > I have a suspicion that we are also seeing Netflix Ultra HD (~1080p) streams > now, whereas the highest we used to get was their HD (720p). I saw > something that Netflix Ultra HD only worked with a Koby Lake CPU and Edge > browser, but so many people are using Smart TVs now, I think streaming on a > PC has become irrelevant. So in addition to the Windows 10 Update problem, > I’m seeing a lot of 6M customers max’d out with what appears to be a single > video stream, while in the past we would see the stream at more like 5M. I > think now it just switches on the fly between HD and Ultra HD, to use all > available bandwidth. And then the Windows 10 Update traffic which sends > 180% of the rate limit, plus people get frustrated and turn off the PC, so > it never finishes the download. > > > > From: Af [mailto:[email protected]] On Behalf Of Mark Radabaugh > Sent: Friday, November 25, 2016 12:30 PM > To: [email protected] > Subject: Re: [AFMUG] Cambium 450 QoS details? > > > > Ken, > > > > Not entirely an answer to your question, but hopefully useful. The > backstory - I was getting pretty annoyed that our VoIP techs kept claiming > that Canopy QOS didn’t work even though I have proven with packet captures a > number of times that it does work, so we set up a test lab to see how High > Priority and standard traffic rate limits interact. End result was that I > was only partially correct and it took a little more tweaking to the > configuration than I thought. The Canopy manual is correct - but you have > to read the fine print and it’s not all in one place if I recall correctly. > > > > Anyway - this i what works for our VoIP customers using QOS with a standard > and voice vlan: > > > > For All Dual-VLAN (Internet + VoIP) Juniper SRX Deployments: > > > > On the SM.... > > > > 0. Set "Sustained [Uplink|Downlink] Data Rate" (Kbit/s) based on the > customer's service plan (i.e. 6 down / 1.5 up) > > > > 1. Set "[Uplink|Downlink] Burst Allocation" (Kbit) to a value equal to > "Sustained [Uplink|Downlink] Data Rate" > > > > 2. Set "Max Burst [Uplink|Downlink] Data Rate" (Kbit/s) to a value twice > that of the "Sustained [Uplink|Downlink] Data Rate" > > > > 3. Enable "Hi Priority Channel" > > > > 4. Set "Hi Priority [Uplink|Downlink] CIR" to a suitable value. i.e. [(100 * > #phones) + 100] (aka 500 Kbit/s for 4 phones) > > > > 5. Set "Priority Precedence" "802.p then DiffServ" (either one actually > works, but this is belt & suspenders in case the SRX config isn't fully > implemented). > > > > 6. On the SRX, configure both WAN VLAN interfaces (i.e. internet and phone) > with firewall rules to mark traffic as high priority (phone) or low priority > (internet) [interfaces -> IFNAME -> UNITNUM -> family inet -> filter] > > > > 7. On the SRX, enable rewriting (dscp and 802.1p) on both of the WAN > interfaces [class-of-service -> interfaces] > > > > This config is valid and tested with 80 Mb/s of simulated upload traffic, > 500 Kb/s of simulated VoIP traffic, on an SM with only a 1.5 Mb/s upload > cap. No lattency/jitter/packetloss on the simulated VoIP traffic. Only 15-30 > ms latency on the internet upload (not bad, considering the SM is forced to > discard 95% of the traffic it's getting over the wire). > > > > We didn’t analyze the traffic to see how the drops were handled on the > SM->AP side. > > Mark > > > > > > On Nov 25, 2016, at 12:56 PM, Ken Hohhof <[email protected]> wrote: > > > > I also posted this question on the Cambium Community forum, but maybe this > is well known and someone here can answer for me, or point me to the info. > > > > How exactly does Cambium 450 QoS work? Does it buffer traffic and shape it > to the configured rate, or does it do traffic policing? And if it buffers, > what is the buffer size? Is it just a global buffer pool whatever memory is > available, or a fixed buffer per SM/LUID? If it does policing (and even > shapers have to discard traffic if the buffer overflows), does it discard > whole packets (PFIFO instead of BFIFO)? > >
