After switching AP mode from Flexible to GPS sync 75/25 i've discovered that 5min CPU usage become quite strange.
Flexible mode idle cpu usage is 25-30% and 75/25 mode usage is 5-7% with strange 100% peaks.
Tested on two sectors EPMP1000 GPS 5GHz firmware 2.5.
Idle usage on unsync connectorised EPMP1000 on AP mode are the same in flexible mode, but in 75/25 mode usage is 10-15% without any peaks. Trafic on both AP about 2-5mbit, 6 SM connected to GPS AP and 3 to Unsync.
Can anyone explain big differences in cpu usage on flexible and 75/25 mode on GPS AP?
With regard to Flexible mode using more CPU than 75/25 mode, I believe what is happening is that Flexible mode allocates frame sizes on the DL and UL anywhere from 1 - 5 msec. Thus, an empty link in both directions might have 1 msec DL size and 1 msec UL size whereas a link loaded in both directions might have 5 msec DL and 5 msec UL. Since your links are mostly idle the frame sizes are smaller. Creating a frame takes CPU time and since most of the frames are small in size, this consumes CPU.
When 75/25 is used, 3.75 msec is used for the DL and 1.25 msec is used for the UL. The frames will have more idle time and are on average bigger in size than the Flexible mode frame size. Therefore, CPU utilization is lower.
This means that everything is good, because in Flexible mode the increased number of frames with smaller times keeps your latency lower which is what you want. If the throughput on the link increases, it probably does not translate into much of a change in CPU.
Now why you are having spikes in CPU usage for 75/25 mode, I am not sure bout this. I will bring this topic up with the software team and get back to you.
We have 22 ePMP gps-sync APs (2.4 and 5Ghz) using the the little GPS puck and every one of them exhibit this behavior. We assumed it was a graphing glitch since we haven't noticed any performance issues. We also have one we just set up and even though it doesn't have a single SM connected to it yet it still has these spikes every few minutes except it only spikes to around 92% instead of 100%.
Hello Cambium Support is possible to do something with this case? In our monitoring all CPU peaks ( caused probably by GPS ) they have impact to time response (ping) for our customers. Thank you for any progress ....
This is still occuring in ePMP release 3.4.1 when polling via SNMP. Yes, we realize the number needs to be divided by 10 to get the usages with one decimal point. We will get spikes where the raw return values are 1000 for a number of minutes.
This can be on AP's with or without clients for us.