I suspect that I am having a negative flow control interaction between a Force 300, UBNT EP6 configured as a switch, and some customer's mikrotik routers.
Is it in any way possible to disable flow control on ePMP 3000 gear for testing this theory? This may also be a switch buffer issue, but taking the ePMP link out of the equation makes the issue go away.
This is easy to reproduce with the following components -
3000 AP to f300 16 cpe. DS9 up and down, link distance is about 500'
f300 connected to ep-r6 configured purely as a switch (1G ethernet)
two customers with Mikrotik HAP lite routers wired to switch (100M ethernet)
Wireless link test between AP and STA good for 120x25+. iperf test between AP switch and STA switch is equivilent. Throughput to customer mikrotik routers never exceeds 55~60Mbps. Throughput between mikrotik routers connected to the switch is 100Mbps as expected. Plugging a laptop into the switch directly (1Gbps ethernet) gets expected speeds.
I appreciate you doing that, RayS. Unfortunately, it seems like everytime you need to chime in with this on forum posts, it goes no further.
I can state that during further testing, when the epmp3000 gear is removed from the equation everthing performs as expected. This is either a flow control or buffer memory issue in this specific case. I have other epmp 3000 ap units servicing the same cpe and router gear with no speed issues. The issue here is the switch on the lan side of the f300 unit.
We had this same issue not too long ago. F300-25 feeding a cheap 5 port 1gig Netgear switch, which then fed 2 R190W routers. Install guys called me and said they could only get 20-30Mbps through either of the routers when router/s were connected to switch. Connect pc with 1gig port to switch getting 110-120Mbps. Connect either of the R190W's directly to POE brick and getting 90-95Mbps down via lan port on router/s.
I went out to location with a hap lite, hap AC 2, and 6 port Netonix switch.
I first connected the hap AC 2 (1G ports) to Netgear switch and using Tik built in bandwidth test or online speed tests was able to reach over 100Mbps download. Then connected the hap lite (100Mbps ports) and was only getting 20-35Mbps. like the R190W's.
Replaced Netear with the Netonix 6 port, and connected both R190W's to Netonix. Was able to get full speeds via either R190.
Moral of story is replace your switch with something like a Netonix and your problem goes away (or replace your 100Mbps routers with 1g routers).
Btw, our Netonix switches does not show that any of the 3000/300 radios have flow control active. We have connected the 3000, 300-25, 300-16, and 300 CSM to them and none have shown flow control active when we activated FC in the switch.
I am not sure that this is exactly what you experiencing, but let me share what I saw before.
ePMP2000 or ePMP3000 are TDD based systems. Meaning that they are receiving and sending data in fixed frames. The frame size is 5ms. ePMP will receive the portion of data and send it out with the speed of ethernet port(1 Gigabit). It will create a short burst of packet at the almost Gigabit speed.
Coming to the switch those packets has to be switched from Gigabit to 100Mbit port by the switching fabric. Different switch chipset vendors are doing it differently, but all of them invlove buffers to do this. Buffers sizes can be different from switch to switch. If the buffer is not enough to hold the burst of packets some of them might be lost that will lead to TCP performance degradation.
Enabling or Disabling Flow control on any of the equipment will not help with this issue, because it will not prevent the packet loss.
As recommended by CWB it makes sense to change the switch or have the same wired speed across your network(eg Gigabit).