Maximum burst speeds with PPPoE enabled

Overall, we find the PMP450 to be a great platform. However, our main local competitor is using an LTE TDD based platform from Huawei and speed tests on the LTE platform do seem to be giving better maximum burst speeds than we can currently get on the PMP450 with PPPoE enabled. The speeds are about the same on both platforms with PPPoE turned off on the PMP450. We have done a lot of testing related to this issue to filter out the variabilty associated with different signal strengths and related modulation rates, sector load, backhaul capacity, router PPPoE handling, using different QoS parameters etc. We have also tested with the SM in Bridge mode, where the client router is the PPPoE client, and we see similar results.

With PPPoE enabled on the SM in NAT mode or on a cient router with the SM in Bridge mode we can not reliably get the download burst speeds above about 45mbps and usually it is a lot lower than this. We use a 75% / 25% download / upload split on the frame allocation and using 30MHz channels does or improving the signal strength does not make any significant difference to the peak speed attainable. Running the built in link test directly from the AP to the SM will typically show the maxmium speed to be much higher, 65mbps or better if the signal is good. Running speed tests over the same link when not using PPPoE also resolves the performance problem and we see maximum burst speeds similar to what the link test would suggest is attainable on the link.

The real commercial problem for us is that customers are very focused on the the results they get from online speedtests. Of course we all know that these are not that reliable, but they do give an indication of performance and unfortunately for operators like us all that a customer will notice is the headline speed that their connection can achieve. I have had more than one customer say to me that their neighbor can get 65mbps on a connection, supplied by our LTE competitor, while our customer can only get 45mbps using our connection.

It would be tempting to just not use PPPoE at all. However, we made the decision to use PPPoE at the start to help control access to the network and to help preserve our allocated public IPv4 address space. Our conclusion is that there is some packet processing issue in the PMP450 platform when dealing with PPPoE encapsulated data. We would expect a small overhead with PPPoE, given that there are few extra bytes in the header resulting in a slightly smaller MSS available for TCP, but not such a big performance hit for uisng PPPoE.

We are being pushed to offer every increasing speeds to our customers and we seem to be hitting some kind of limit on the PMP450 when using PPPoE. So I have two questions I would like to ask:

1: Have any other operators noticed a similar issue or found a way to alleviate the issue?

2: Do Cambium have any plans to look at this issue and improve the performance handling of PPPoE encapsulated data?

Thanks for any feedback or suggestions.

Question - if you have 2 SMs, each bridged to a client router running PPPoE, if you run a concurrent speed test on both PPPoE clients do you only get 45Mbps total between the two units, or are you able to hit the full aggregate amount available on the AP? Just curious if the bottleneck is present at the SM level or the AP level (or both). 

We have not tested that precise setup in a controlled way. However, we suspect the bottleneck may be on the SM, as we often see the speed through an AP burst much higher than 45mbps when all the SMs on the AP have PPPoE enabled.

What software version are you running? I recall reading some  things about limited throughput despite  the Link Capacity Tests.

We run PPPoE but it's all done on subscriber routers. All of our SMs are bridges.

All devices are running at V14.2.1 with a mix of bridged and NATed SMs.

I have noticed this issue for years and cambium can’t seem to pinpoint it. We do not use pppoe. In both a production and lab environment actual speedtests are about 30 to 40 percent less than a 10 second capacity test. The 450 3.65 platform seems to do better with the 5ghz 450 and 450i performing the worst. Our best test in the lab was 65mbps for about 10 seconds with all qos settings at their max. These units were tested operating as bridges. We put epmp next to the 450i and it just destroyed the 450i in maximum throughput. I have opened many tickets, constantly complain to reps and nothing gets past the discovery stages. It is almost like there is some invisible hard limit. We have tested 10, 20 and 30mhz channels and only see about a 10 to 20 percent increase in throughput between the channel widths. I have a 3.6 450, 20mhz channel, 45 subs that hits 100 percent frame utilization at about 55mbps. I have a 450i 5ghz, 30mhz channel hitting 100 percent frame utilization at about 65mbps with about 60 subs on it. My point is there doesn’t seem to be any consistency in the behavior between platforms, conditions or configuration. The only consistency is under performance. We are looking more at LTE. Yes, we have adjusted control slots in both directions with little change. I will also note that with the maximum burst settings maxed we should test that 65mbps for about a minute. In the lab it tested 65mbps for 10 seconds and then dropped to well below sustained at around 13mbps. In production we typically get 60 percent of what a capacity test suggests. We have been through noise, errors and other factors. Drop an epmp or ubnt next to the 450 and the flood gates open.

1 Like

odd the more powerful pmp radios would be trounced by the cheaper epmp line, especially when i'd hope the 2 lines of developers worked together on things like that. Has cambium said anything about it?

Our issue isn't with overall sector throughput but with the maximum speed that can be achieved on a single device assuming that the sector isn't too busy and the radio conditions are good etc. Our experience is that the PMP450 (or PMP450i) will easily outperform ePMP when you load up a sector. Ubnt are not even at the races when sector load and the number of CPEs starts to grow due near field interference and desensing issues caused by unsynchronized adjacent sector transmissions. It would be great to get some feedback from Cambium on the specific PPPoE issue.

Yes about what i expected, but still, sad that the single radio sm performs more poorly vs a epmp sm for single radio performance. 


I'd imagine its a issue with the processing power on the SM itself, i'm really hoping for the next-gen SM they will gather input on concerns, from carriers/wisps before finalizing the next gen sm. 

Though even with that said, the epmp sm's are dirt cheap, so why would they have more processing power than a pmp450 sm, so maybe that isn't the issue.

So I discovered something I find more than just a little aggravating.

Our system is 450i - 450d-200p 

On a SM 450d capped (40mb) connected at -66, the max speed I could get is 20/12 8x/6x MIMO-B clear LOS nothing near the fresnal zone.

For grins (plus I ran out of other ideas) I applied a uncap license and suddenly it was cappable of 60/15 8x/6x MIMO-B.

Without changing anything other than a license from Capped to uncapped I saw performance I should have been able to get close to on the capped version.

It seems that the license cap on the radios is the function of a ratio of max perfomance and not in fact a simple cap for overall bandwidth.


@ColdBlueFire wrote:

So I discovered something I find more than just a little aggravating.

Our system is 450i - 450d-200p 

On a SM 450d capped (40mb) connected at -66, the max speed I could get is 20/12 8x/6x MIMO-B clear LOS nothing near the fresnal zone.

For grins (plus I ran out of other ideas) I applied a uncap license and suddenly it was cappable of 60/15 8x/6x MIMO-B.

Without changing anything other than a license from Capped to uncapped I saw performance I should have been able to get close to on the capped version.

It seems that the license cap on the radios is the function of a ratio of max perfomance and not in fact a simple cap for overall bandwidth.


Very interesting. What Cambium says?

That is not how it is supposed to work... the throughput license should be exactly as ColdBlueFire describes, just a throughput limit.  It does not depend on modulation or anything else.

We did, however, have an issue with license key functionality  that was recently discovered.

ColdBlueFire - If you can PM me, I would like to do a little experiement with you if you're willing...

I am adding to this thread.  We have a very similar issue.  

Client to AP speedtest is 40 megs.  Great signals, etc.  Can test all the way to the core and great speed.  

To be more specific. Particular client can pass 40 megs tcp across a link to another AP and to the local AP.  Bring up PPPoE and throughput drops at least 20-30%.  That 40 meg is now 20-25 meg.

Only thing different in the tests is traffic going over dhcp vs PPPoE.  Happens across the board.  Gear is 450,450i.