Hi All,

I have just setup a link for a customer using a pair of RB411AH, Mikrotik > Mikrotik is doing 80mbps TCP using the test tool however from a windows machine running the iwndows BW tool I only see 40mbps tcp and this is reflecting the internet speed also which is topping out around 45mbps with a 100mbps connection on the other end.

I have check everything I can think of, CPU load on both devices is between 35 - 45% at 45mbps, If I plug in directly on the other end the internet connection is doing roughly 120mbps.

http://dl.dropbox.com/u/8790688/mikrotik.PNG

Any pointers? or suggestions.

Cheers

Cameron

Hi All,

I have just setup a link for a customer using a pair of RB411AH, Mikrotik > Mikrotik is doing 80mbps TCP using the test tool however from a windows machine running the iwndows BW tool I only see 40mbps tcp and this is reflecting the internet speed also which is topping out around 45mbps with a 100mbps connection on the other end.

I have check everything I can think of, CPU load on both devices is between 35 - 45% at 45mbps, If I plug in directly on the other end the internet connection is doing roughly 120mbps.

http://dl.dropbox.com/u/8790688/mikrotik.PNG

Any pointers? or suggestions.

Cheers
Cameron

In Btest window, change TCP Connection Count to 1, and make another test. Put results here, then I can explain you why the results are like you descibed.

So I changed it from the default of 10 to 1 and got a similar result:

http://dl.dropbox.com/u/8790688/Mikrotik1.PNG

and as you can see a similar result for the traffic to the internet: http://www.speedtest.net/result/1841278263.png

What is your CCQ? Signal strength? Have you tried a different channel?

Any firewall or mangle rules? Queues?

So I changed it from the default of 10 to 1 and got a similar result:

http://dl.dropbox.com/u/8790688/Mikrotik1.PNG

and as you can see a similar result for the traffic to the internet: http://www.speedtest.net/result/1841278263.png

Now everything looks like it should. `
TCP traffic needs ACK pacekts. Single session throughput strictly depends on RTT, MTU, and errors in transmission (BER).

Problem is described here: http://www.cost285.itu.edu.tr/tempodoc/ ... otorti.pdf

In wireless links the most important thing is BER. If you have to many errors in your link, it affects session throughput. Programs which use many TCP sessions for testing will show better result, than one session check. Default MT use 20 simultaneous sessions, and web tests use one session for checking.

Hi,

http://dl.dropbox.com/u/8790688/Mikrotik2.PNG

Cheers

Hi,

http://dl.dropbox.com/u/8790688/Mikrotik2.PNG

Cheers

Poor CCQ, bad signals - check antenna system.

Even that is not really clear. Was the link under any load at the time?

Ok so I switched one of the chains around as it was mismatched and tried to align the antennas a little better and now it looks like this: http://dl.dropbox.com/u/8790688/Mikrotik3.PNG

Ok so I switched one of the chains around as it was mismatched and tried to align the antennas a little better and now it looks like this: http://dl.dropbox.com/u/8790688/Mikrotik3.PNG

What antennas do you have there? -59 on one chain, and -72 on second... it looks like antenna problem.

A Pair of http://en.jirous.com/antenna-5ghz/jrc-24-duplex

A Pair of http://en.jirous.com/antenna-5ghz/jrc-24-duplex

Antennas are almost perfect, so it have to be something else. Maybe a card, or pigtail.

Are the antennas well aligned?

Maybe you may check vertical alignment on both sides.

If a dish is a little rotated, the cross polarization attenuation will decrease a lot, and CCQ will reduce on consecuence.

I recommend you use a bubble level to check

If you can find the video, there was a conference in the last MUM that explained this

Regards

MUM last week?

MUM last week?

Yes. If I remember well, that conferenced was exposed by Deltalink

Regards

Hi,

So I havent had a chance to check the antenna at the primary site yet so I switched the AP to send on the good Signal/Condition Chain and recieve on the misaligned chain and the ClientAP the oposite as below:

http://dl.dropbox.com/u/8790688/Chains.PNG

As you can see the Signal and CCQ is now perfect for the recieve chain:

http://dl.dropbox.com/u/8790688/signal.PNG

However we are still only able to on 1 connection get 35 - 45mbps and 4 connections we get 75mbps/100% cpu

http://dl.dropbox.com/u/8790688/1%20stream.PNG
http://dl.dropbox.com/u/8790688/2%20stream.PNG
http://dl.dropbox.com/u/8790688/4%20stream.PNG

So Any ideas how with 100% CCQ and 150mbps on the chain 100% of the time how to achieve a higher single TCP session throughput?

Cheers

Are you running btest into the same AP?

It eats lots of CPU, and that decreases perfomance.

You may use another board in each side, or 2 computers running btest

Regards

Hi

I'm running the test from a windows machine behind the client out a 100mbps connection on the ap side. No routing is being done by the units. CPU load is roughly 33% when doing the external speed test and getting 45-50mbps these results are matched if I do a Btest from windows across the link to the ap yet client to ap gets 80mbps.

Cheers.

If you make btest between computer and same side AP.... do you get 90mbps?

Maybe your ethernet cable is too long.... or it's damaged

Regards

Hi,

So I havent had a chance to check the antenna at the primary site yet so I switched the AP to send on the good Signal/Condition Chain and recieve on the misaligned chain and the ClientAP the oposite as below:

http://dl.dropbox.com/u/8790688/Chains.PNG

As you can see the Signal and CCQ is now perfect for the recieve chain:

http://dl.dropbox.com/u/8790688/signal.PNG

However we are still only able to on 1 connection get 35 - 45mbps and 4 connections we get 75mbps/100% cpu

http://dl.dropbox.com/u/8790688/1%20stream.PNG
http://dl.dropbox.com/u/8790688/2%20stream.PNG
http://dl.dropbox.com/u/8790688/4%20stream.PNG

So Any ideas how with 100% CCQ and 150mbps on the chain 100% of the time how to achieve a higher single TCP session throughput?

Cheers

Remember that TCP i a protocol that use duplex connections. It sends ack, so making one side ccq better can't solve 100% of your the problem.
What can you do then :p ?
Try to set data-rates maualy for one step lower than a max from auto. In example - if your link is now working with max MCS7, set manually rates to MCS1-MCS6.

So after all this time the link is still running at -51 both chains 100% ccq and 300/300 yet a single TCP session can only do 50mbps. Have tried routing, bridging and tunnels and all the same result.