I've got a RB800 loaded with 3 SR-5's. One link in particular is acting quite bizarre in that as long as there is very little to no data flow the throughput seems stable at around 25M-30M as viewed in the wireless tables. Radio signal is in the low 70's with little to no variance. However , once I start pumping any appreciable amount of data across this link , the throughput all but collapses down to double digits. The other end is a brand new HPNs Metal . This has really messed with my head as to why increasing data flow results in drastically diminished throughput . All my other setups are very similar and don't do this. Anyone ever see this behavior and what did you do to correct it. My options are pretty much to change out the board and cards.

it could be power related, i have had a similar issue with R52HN cards, then splitting it over more than 1 RB solved the issue for me, but there could be a lot of things that could impact this all the way from CPU, to interference, You will have to start a testing process, whats the AVG cpu usage on the tik, and whats the power usage?

cpu less than 10%....

Changed the SR-5 to SHPn...no change , no stability , rates all over the place. The other end of this link is a brand new SHPn Metal. Arbitrary peaks were impressive though , nearly 70M , it just won't sustain. Settles around 10M but also drops to less than 1M. It's just all over the place. This box has been in service for over 2 years with no substantial issue. Just puzzles me as to why it just started this a couple weeks ago. Interference? I've put this thing on every available frequency possible and it does the same thing no matter. Bought an RB-435 , my next move is to just replace the whole RB-800 board , just out of ideas.

Mikrotik , how is the P throughput figure arrived at in wireless tables registration? As I've stated , little to no data transfer results in high throughput figures. Executing a flood ping or ping speed test , the throughput drops off substantially ,as much or in excess of 90% of pre-test execution.Still trying to figure this one out....two other radio card links in the same unit are unaffected.

I've got a RB800 loaded with 3 SR-5's. One link in particular is acting quite bizarre in that as long as there is very little to no data flow the throughput seems stable at around 25M-30M as viewed in the wireless tables. Radio signal is in the low 70's with little to no variance. However , once I start pumping any appreciable amount of data across this link , the throughput all but collapses down to double digits. The other end is a brand new HPNs Metal . This has really messed with my head as to why increasing data flow results in drastically diminished throughput . All my other setups are very similar and don't do this. Anyone ever see this behavior and what did you do to correct it. My options are pretty much to change out the board and cards.

You may have two issues here or perhaps one?
(1) With 3 radio cards onto one MT device can cause interference between the radio cards and the proof is as you say when data increases the throughput collapses, I have a simple policy, 1 radio card = 1 Routerboard also integrated Antenna’s gets rid of RF leads (LMR400, etc.) so the losses from radio card to antenna are reduced, also in your present setup if the board goes down you lose 3 radio cards and customers connected to those cards,
(2) Co- location interference pickup when one antenna picks up signals from the other antennas on the mast, have you done a signal scan from each antenna and checked if the results show signals below -30, I had problems with antennas on a mast with this but solved it with RF screening.http://forum.mikrotik.com/viewtopic.php ... &hilit=+rx
(3) When you say “All my other setups are very similar and don't do this….” Are they using exactly the same hardware, same distance apart + position on mast, same frequencies, etc. if not then they cannot when RF is involved be considered “RF similar” but appearance are.

I'm beginning to believe this is routeros software related , I took this screenshot of a ping speed test over this link. Look at the average...impossible. This is not photoshopped or any other image manipulation. It is a direct screenshot of an active test.