This is something we rely on quite a bit to test for available 'real usable' capacity but 99% of the time we've been using it 1 hop away, either direct router-router connections or across PPPoE connections at customer sites, and for this its fairly accurate especially with UDP
However i've been testing multiple hopes lately and notice its gone to shit and will report completely wrong figures. I.e. RouterA->RouterB shows ~200mbit/s. RouterB->RouterC shows 150mbit/s. So you'd expect RouterA->RouterC to show ~150mb/s but it only shows ~60mb/s

Has BTest gotten more broken in later firmware releases or is this a longstanding issue that I just havn't noticed until now?
No, iperf is not a suitable replacement, i'm aware running tests 'on' a router aren't as accurate due to the extra CPU processing, but I watch this and as long as a CPU core is not being saturated I can typically rely on the results, otherwise UDP adds negligible load. And from a real world perspective I'm not going to drive out to sites all over our network to run bandwidth tests sporadically, its just not feasible
So we need a way to check bandwidth in our network properly. And we'll be moving our customer base off PPPoE to DHCP and a part of this reason is to ease troubleshooting of slowdown by being able to test at each hop inside our network rather than just the entire tunnel end to end. But if BTest is broken after 1 hop, this is actually worse

My thought, and a lot of guess work ...

If these are wifi links, and the link is RouterA-RouterB-RouterC then I don't think Btest is broken, but your actual throughput is 60 Mbps.

If the transmission is in the same ether cel, same channel, then you have co-channel interference. There is only one router sending.
In the ideal world (time domain multiplexing) then the theoretical maximum speed is (1/200+1/150 = 7/600 = 1/85) or 85 Mbps.
But there is contention and clear channel testing, and collisions (and retransmit) and radio send/receive transitions .... so 60 Mbps is a very nice result.

Adjacent channel interference will be worse.
Cross channel connection (different channels, no interference) could give you up to 150 Mbps if no congestion occurs.
(TCP will avoid congestion and slow down (too much) , UDP will create congestion)

No, doing independent tests from A to B and B to C shows roughly the actual bandwidth available. Testing A to C shows way lower
It's not interference, completely different (80ghz one side, 5ghz the other). Radios are not pointing in the same direction

A to C should be seeing 150 but it doesn't

I tested this in some other areas of the network as well where there's at least 350mb/s available on all paths, was also seeing far lower results when going more than 1 hop, down around 80

OK, no interference then.

Can you tell some more about the devices used, specifically the RouterB and how the connection/config is between the 60GHz and 5 GHz interfaces. One device/ two devices?
Is B routing or bridging ? etc etc

(I have done Btest over long paths, never seen degradation).

Ok its not the btest program, it's something funky going on with the RB3011's in our network. They can do good performance only when passing packets to a directly connected neighbor, but if they have to go more than 1 hop it all turns to shit. Even if packets are being passed 'through' a RB3011 and not 'from/to' one its still total shit
VPLS tunnels are not affected, but normal IP routing is
Thread will continue here viewtopic.php?f=2&t=159498