As you may be able to see, at 20:00, the CPU usage hits 100% when downlink traffic crests 60Mbps, and at that time, Frame Utilization flatlines at 80%. Before and after that period of high cpu load and flatline of frame utilization, you can see that the frame utlization spikes up slightly. It's almost like the AP was maxed out, even though it was hitting only 80% frame utilization.
Based on that flatline of Frame Utilization at 80%, lining up with the 100% CPU load, I wonder whether AP is actually hindered?
Ultimately what I trying to determine is whether the customers are experiencing slow speeds when the CPU hits 100%, even when the frame utilization has not. From what I'm seeing on my graph, I feel that there are indeed slow speeds when that happens, but I'm hoping I'm wrong...
Yes, your customers do see speed limits at 100% due to packet per second processing limits of 450. The solution is to upgrade to a 450i which has a much faster fpga.
We’ve done this regularly and see a jump in performance.
Example attached showing an upgrade from 450 to 450i about 3 weeks ago. Notice the jump to ~ 80mbits max from ~ 60mbits.
Yes, your customers do see speed limits at 100% due to packet per second processing limits of 450. The solution is to upgrade to a 450i which has a much faster fpga.
We've noticed this with the original 3.65 450 AP's and upgrading them to 450i's helped... unfortunatly, it appears as though the OP is using 450 on 2.4GHz... which does not have a 450i radio to upgrade to.
You can squeeze a bit more performance out of it by using 5ms frames, auto contention slots, and not using extra services like syslog, protocol filtering and cnMaestro agent.
Lastly, you could move to ePMP on 2.4GHz which IMHO performs better and is obviously cheaper then PMP450.
Unfortunately your fear and Eric is right. When the CPU hits 100% it cannot keep up with the packet load and will drop some packets even though there is actually room on the RF pipe.
The PMP 450i AP has a more powerful CPU, which is why it was developed. But as mentioned, there is no 450i in 2.4 GHz.
I agree if you limit the extra services and keep web page refresh off, limit SNMP monitoring, you can keep CPU a little less busy, it will help. Also of course you could add a AP and further sectorize, but I know that costs money.
We also continually try to optimize the packet processing. I know of one small improvement coming in 16.2 BETA-6 and another one we are working on for a future release. These will apply to the whole 450 product line, including 450 AP.
I'm curious why you would say that the ePMP3000 performs better? As you astutely pointed out, a 450i isn't an option for me. Anyway, I was under the impression that the PMP450 was a better radio in all aspects, other than of course the 4x4 MIMO of the ePMP.
This particular AP has 31 subscribers, many subscribe to our 25 Mbps package, and it's pretty efficient in terms of the quality of subscriber links. We have already made some of the changes suggested, such as 5ms frames, and autocontention. I didn't think to disable Auto-refresh, cnMaestro or LibreNMS, so perhaps theres something to be gained, although I'm reluctant to think those last three will make much of a difference, but I'll certainly consider it.
But again, I'm very reluctant to think changing from the PMP radio to an ePMP is the solution, however I'm very open minded and curious to hear why you feel it is.
I'm curious why you would say that the ePMP3000 performs better? As you astutely pointed out, a 450i isn't an option for me. Anyway, I was under the impression that the PMP450 was a better radio in all aspects, other than of course the 4x4 MIMO of the ePMP.
I didn't say e3k performs better or suggest you move to it... the radio that you have in the graph is a PMP450 2.4GHz AP... I suggested that replace it with e1k for 2.4GHz. You can actually sync between the two systems to help ease the transition. We prefer e1k over PMP450 for 2.4GHz because it's cheap, has higher TX power, don't have to use a reflector dish, seems to handle higher PPS and throughput, etc... the list goes on...
The major advantage that ePMP has over the early PMP450 radios is that it uses a mature 802.11n WiFi SOC which has many RF, networking, services, etc. accelerated by being baked in silicon... vs. PMP450's FPGA which on the older models is pretty slow by today's standards and when running later firmware builds with lots of features/services enabled, will eat up the FPGA's processing time.
Apologies for my confusion Eric, I didn't realize the only 2.4 option for ePMP was the 1000.
Thank you for the clarification and further explaination.
The 2.4 PMP450 is definitely starting to show it's age, that is for sure. Stiil works great for us in so many places, until we really load it up with heavy users as the evidence shows. Anyway, I greatly appreciate the advice, and will certainly be considering the ePMP1000 for 2.4 deployments in the future.
