Throughput versus speed tests versus link tests discrepancies

I could say, at least with the network I work with, there are a lot of times that I'm troubleshooting slow speeds. And I'm posting this in epmp, because it just so happens to be the last radio I've had this particular set of circumstances happen.

Someone calls about slow speeds, I have them bypass router, go directly into POE. Throughput bounces up and down, but ultimately ends up at 10mbps. Speed test on their end is 7mbps. I run a link test, and I see 16mbps. This particular customer is provisoned at 15/3. 

And for all intents and purposes of possibly helping me out with this, since they are an issue I personally struggle with more than anything, and understanding how to deal with slow speeds would be awesome, I'll post some screen shots of their stats if you'd be ever so kind as to help me..But mainly just interested in the descrepency.

All I can think of is, is that throughput doesn't reflect disk writing, and link tests don't involve hops to a server.  

The images show a link that should be good for much more than 10Mbps.  The nearly 5% MCS1 on downlink may be a sign of local interference though.  It may also just be a sign that the connection has been sitting idle for a while.

How is the AP fed?  Is there sufficient backhaul capacity?  Are there any switches between backhaul and AP choking on pause frames or with small buffers?  There may be a lot more to this than just the final hop from AP to CPE.

Keep in mind that the link test in the radio GUI is UDP and may not accurately reflect TCP throughput.


You can also try fixing the maximum MCS to MCS12 for this subscriber. It will slightly decrease latency and you will see better TCP performance.


Tower is is fed from two PTP 800s, runng around 230mbps, (460 duplex) from two towers in the path group into a netonix, into the ring. This is also a very busy site. (Not our busiest, but comparatively) Ring of 7 pmp320s, ring of 4 epmp2000's, 6 M5 rockets, 5 M2 rockets. We're in central Florida, so we got every kind of interference imaginable, so we need all the spectrum. 

I'm pretty new at this, I mean this is my second month working at this job. I sent it to the 3rd shift guy to take a look at, I'll share what he said this SM is on "HC-15", so you sort of understand what he's talking about. 

Pulled up another customers SM and it sees HC-15 South AND WWT South.

WWT South is 5230/40 Which gives it the spectrum 5210-5250

HC-15 South is 5190/40 Which gives it the spectrum 5170-5210

WWT South and HC-15 South are overlapping on channel 5210.

At 20mhz, what could that AP serve at MCS12? Because I had subscribers that are set to 25/5, do you have a chart? I found one that works with the modulation rates for pmp320s but nothing past 10mhz. Also, I can only set the uplink MCS on the SM, I found a setting to set downlink, but it's on the AP. 

DirkDeadeye wrote: At 20mhz, what could that AP serve at MCS12? 

I think he was suggesting fixing the SM's uplink at MCS12, not doing anything to the AP.  At least I think that was his suggestion?  :)

This particular SM's uplink is most of the time failing at MCS 13-15 anyway, which means that it needs to fail and retry back to MCS12, so that'll add latency and overhead. I think he was suggesting changing this particular SM to MCS12 which would prevent it from bothering to try 13, 14 & 15 for it's uplink data, and seeing if that helped the situation. 


@DirkDeadeye wrote: you have a chart? 

The charts you ask about are in the ePMP user manual/guide. 


Thanks for better explanation. This is exactly what I meant.


I'm the 3rd shift tech. The MCS rates are jumping all over because there is an overlap between this customers AP and another AP located elsewhere. Just so happens that the two APs converge on the same frequency when they are set at 40 width/5210.

I saw this by logging into another SM and they saw both of these AP's despite their distance from each-other which is quite a large distance.

I may be able to tweak the transmission output power or use some down-tilt on the offending AP so that SM's can no longer see 2 different AP's from 2 different tower locations.