I’d like to change our wireless link (x86 PC based 5ghz-turbo mode mikrotik) to a faster one. At firs I was looking around the nstreme2, but there were some problems when trying on the desk-top.
1, tried with 3 sets of wireless cards
2, with two routerboard, and two PC-s and every combinations of them
3, with 5G band UP-DOWN-link, and 5G UP and 2,4G downlink (to avoid any kinds of interference)
The problem was: The NS2 brings up normally, when I was ping the other router, it’s OK, but when started a bandwith test, the link dropped, and only after a reboot connected again.
On an other site I was read, the Bonding of interfaces is a better solution for this, but:
I was set up the two links, and can nicely ping the other side, with 0ms response times. I create the EoIP tunnels, bonding them into 1 “bonding” interface. And at this point startint to cry, 'cause the answers for the ping requests has 30-70-80 ms times, WITHOUT ANY TRAFFIC!
At now it works with single 5ghz turbo card, at the rate of 23Mbits/sec, but getting to be isn’t enough, so it must be replaced with a 35-40M capable link.
Has anybody ideas to connect two sites with a more faster link (11km)
Well first off - 11km, you are going to need some high gain antennas to reach the signal levels required to push 40Mbps of data across a wireless link. You’ll need about -60db to achieve that level of throughput.
Minimum cpu for this would be at least an RB333, a ‘pc’ would be faster if you can set them up on both ends without long RF cables.
You can use 5Ghz on both radios - just don’t make the channels adjacent to each other… Use 5745 and 5824 - or 5745 and 5785 plenty of seperation there!
I have used Nstreme dual on a couple of links. It is fast but I prefer using single Nstreme links and then bonding them together. You get two benefits, double your transfer speed of the lowest pair, two - if a link fails (using rr mode with arp monitor) you don’t lose the entire link - you will maintain connectivity - just a lower overall throughput. I have acouple of these setup w/3 5Ghz cards each - they are FAST. Ping times can be a little deceptive if there is no traffic or very light traffic, but once you start putting about a 10% load on this ping times will improve dramatically.
What are you using for the bandwidth test - the RB or ROS test itself - if so you’ll run out of CPU time before you reach max throughput - be sure to connect an actual pc at each end and then do a bandwidth test from pc to pc through the wireless link.
I was playing with them -two ROS on x86 PC- today. The nstreme2 was succeeded with 36/36M (signal: -29; -31) and 55/20M assimmetrically, that is a little bit more than I wanted before.
So:
The nstreme2 was working fine.
I was set up an interface bonding. When the cards was in nstreme mode the troughput (with the signal levels mentioned before) was 36M in only one direction. With only one nstreme this is somewhere 40-45M in one direction. Conclusion: something is wrong.
Without nstreme mode this is about 22-23M. I can see, the one half of the traffic is on the first, the other half is on the another link.
Have you got any idea? Has anybody experience, what a hardware needed for such a traffic?
I guess I should post some of my real world results.
With RB333, ROS 3.10rc, Nstreme dual, sig levels -55 and -61db, 2 miles apart, 5ghz band, turbo mode, csma off, nstreme polling, on I was able to manage 50+ mbps TCP consistant.
Same setup as above 1.8 miles apart, two single Nstreme radios in each unit, -53 and -59db, used EoIP tunnels, bonding rr and arp monitor, 50+ mbps TCP. Single link (turned one off) 30+ mbps TCP.
The above results were as fast as I could push the CPUs attached at each end of the wireless links - the CPUs hit 100% - Windows XP
May I ask you to send me the config of one of two pairs. 'Cause I can’t set up the bonding to work faster then 22-23Megs if bonding turned on. With one 5G turbo link it works 1,5 times faster. I can’t understand. If you could send it to me, write in forum, and I’ll send you an email that you can reply on.
zhkako -
Sorry - I am not going to post my config here on the forum. My email address is in my signature below.
A couple things come to mind as I read you post over again… Right now you have signal levels at -29 and -31db Your radios are TOO CLOSE. At those signal levels it is just like having someone standing next to you yelling in your ear - you can hear them but you can’t understand what they are saying… So - you need to attenuate the signal some, it needs to -50 or so for your testing to be more accurate.
You may want to try two seperate Nstreme channels. Set your system up as a dual gateway. See what you can get for bandwidth there before proceeding on to bonding. Make sure you are using either 2.9.49 or 3.10rc in your systems.
Now on nstreme you need to remember that pings times are ‘bad’ until you start getting data to flow across the links - I found that at about 5 - 10% minimum of link capacity seems to be the ‘magic’ number for getting good ping times. It can also help if you set your ping packet size to 1500 bytes, this will get it out on the wireless much faster that the standard 32 byte ping request.
Make sure the radio channels are a couple of channels apart - I like using them in different bands if you will, e.g., 5.180Ghz and 5.745Ghz (US we have the channels available - just have to watch the power out on the lower channels).
Turn on Nstreme polling, turn off csma, you may have to go in to terminal mode to make sure csma is off. I don’t use any framer policy.
Start with standard channels to make sure everything works, go to turbo if you need to…more radio spectrum efficient if you do not need turbo…