We have epmp1000 Access point with more than 100 subscribers. We use flexible Downlink/Uplink Ratio and 40 mhz channel width. When the AP is under the load of 50-60 mbit/s we have high jitter and latency in our CPEs (jitter -200ms and latency 60-100ms . Max capacity of this AP is about 80-90 mbit/s. There are no any other devices working near in that frequency. Is it normal? Are there any ways how to reduce it? Firmware version of all of the devices is 2.3.4
Hi Alexander_1. We currently don't have Cambium radios deployed, and we are still currently evaluating Cambium vs Ubiquiti vs others. However, I'm very interested in what you said here...
Just to be clear - you are saying that you have a single Cambium ePMP Access Point, with 100+ clients attached, running 80-90 Megabit throughput at peak time, and that (now) your latency is low across the board? If so, that's very impressive I think. :) I'd love to hear more details - whatever you have time to share, please. :)
Could you elaborate on your fix? You mention limiting the MCS on the client side, I assume you were able to do this by using firmware 2.4? How did you decide which SM's to limit? What is your jitter and latency like now?
How did you limit the MCS on the client side? And how did the exactly help your situation?
Thanks,
Curt
You have to use firmware 2.4+, and on the SM (and AP) you can set the maximum upload modulation. By looking at the wireless statistics on the SM you can find out the %'s that each modulation is being used. It appears as though you might be able to reduce jitter by setting the maximum modulation on the SM to whatever the highest % successful modulation rate is... e.g. if MCS15 is at 10%, and MCS14 is at 40%, setting the highest modulation rate to MCS14 keeps the SM from trying to send packets at MCS15 which would cause either dropped packets OR re-transmits, which would cause jitter.
How did you limit the MCS on the client side? And how did the exactly help your situation?
Thanks,
Curt
You have to use firmware 2.4+, and on the SM (and AP) you can set the maximum upload modulation. By looking at the wireless statistics on the SM you can find out the %'s that each modulation is being used. It appears as though you might be able to reduce jitter by setting the maximum modulation on the SM to whatever the highest % successful modulation rate is... e.g. if MCS15 is at 10%, and MCS14 is at 40%, setting the highest modulation rate to MCS14 keeps the SM from trying to send packets at MCS15 which would cause either dropped packets OR re-transmits, which would cause jitter.
Hello,
Just wanted to chime in here from Cambium perspective. Locking the MCS is actually the correct course of action without seeing much more background data. One of the key statistics to keep an eye on is the retransmission packets as compared to transmitted packets. If the % is high then that means there is some type interference forcing retransmission of packets. While our algorithm handles these retransmissions without any adverse effect to TCP it does increase overalll latency and jitter. In this case our Rate Adapt algorithm is being a bit too aggressive in allowing more retries to get to a higher modulation which in conjunction with the fact that we have 100 subs is increasing overall latency and jitter. In the above example, it would be good to do the following -
From the AP perspective - check DL retransmission vs total packets transmitted
If this is high, consider lowering the max MCS a level below what the system is trying to achieve
Check link quality and capacity for all SMs and see which ones maybe suffering
Check those SMs and look for their UL retransmit rates - consider locking that particular SMs UL max MCS to one level below.
All in all, great that you have a 100 SMs in your system. We have some features in development that will help the latency cause even more. Around end of July we will 2.5ms frame which will reduce latency and retransmission costs. Later in the year, we will introduce a concept of contention window for UL bandwidth request that will cut some time down for such a high number of subscribers.