Multiple frequencies and bad performances

RouterOS Wireless Networking
1

Hi,
I’m doing some tests since 2 weeks now, to improve bandwith on my networks… maybe my questions are quite “amateur” :blush:

I’ve installed two EPIA 1ghz mainboards with 2 routerboard 14 and on each, 3 Winstrom CM9

I can easly get 40 Mbps TCP throughput on 5 Ghz-turbo mode, and 22 Mbps on 2.4 Ghz-g
Frequency are quite clean and low usage…

My tests are based on the 3 non overlapping channels of 2.4 Ghz…

SSID: CM1, frequency 2412
SSID: CM3, frequency 2457

I have 22 Mbps bandwith traffic running on 2412 and when I start a second transfer on 2457, thoughput is falling down on both channels arround 1 Mbps…, but registration stay at 54 Mbps and signal over -50

The only way I’ve found to have High bandwith on multiple interfaces, is one link in 5 Ghz and another in 2.4Ghz…
i also tried with 5Mhz and 10Mhz channels, I have the same results…

Is there a transmit power problem ? pretubations due to the physical proximity of the cards ?
Anybody experience the same things ?

Thanks, Julien
http://www.wifx.net

Bandwidth problem when using 3 Wireless interfaceses
Near channel interferance with 2 cards on same router board
2

Hopefully someone can help you with this… this is quite interesting.

3

For me, it’s weird.

I have RB230 with RB14 using 4 CM9/R52 cards, and it can work together. Yes, it’s not 22 mbps each, but I got better than 1 mbps each with All card using 2.4 GHz.

4

Yes, really weird,

But with staros I have the same result, I was thinking it was from MKT but not… Did anybody do some performance tests running on multiple frequencies at the same time ?
I also tried with a couple of hardware, with 1 minipci per adapter, pentium, celeron, WRAP, etc, etc…
If I have two interfaces running on the same channel, it’s share bps better than with 2412 and 2457 for example…

Also doing the same thing with 5180 and 5745, and it’s something about 600 Mhz between… It’s not 1 Mbps, but divide the throughput by 2…
What kind of CM9 card are U using ? Winstrom ? Is there another model ?

Thanks
Julien

5

Hello,

As mentionned on this post, I had bad performance with multiple channels, regardless of the non-overlapping freqnencies…

To increase bandwith on WDS chains I’ve made some tests with “/sys int wir brust-time”.
And apparently 1500 [us] is the best setting, for UDP and TCP…

But it also solved the problem I had with multiple freqnencies…
Does anybody has informations on this burst-time ? I’ve found quite a few infos on the subject…

Thanks,
Julien

6

This is the ‘friendly’ backoff function of the CDMA protocol that uses in wireless 802.11a,b,g standards.

If the radios are colocated ie. 2 radios in one box both operating of the 2.4ghz frequencies you need as much seperation as possible both by frequency and by antenna.

I have documented this case quite throughly and I will summarize my findings:

My test setup was 4 RB532 all in pacwireless DCE Cases
Antennas used for testing were small 5 dBi antennas, they could be attached both to the box directly or on a LMR400 cable to increase antenna seperation by about 6 ft.
Each box had 2 radios in them (either SR2, CM9 or EMP-8602), more on the different radios later.
All tests were started with EMP-8602 cards with tx-power set to 1 for all speeds. The EMP-8602 cards are 6th gen, and can listen and tx on 5mhz, and 10mhz channels. The SR2 and CM9 always listen at 20mhz by can tx on 5mhz. To my surprise I could find NO difference between the new EMP-8602s and the SR2/CM9 with respect to this issue.

The birds eye view was silimar to this

Box 1 <6ft> Box 3
| |
| | 20ft
| |
Box 2 <6ft> Box 4

Again to summarize if the radios are in seperate boxes you need 30 mhz of seperation to achieve 90% of max speed on both links (ie. box2 tx to box1. and box4 tx to box3 at the same time). To achieve full speed 35 mhz of seperation was required.

If however the radios were in the same box (ie Box2 had 2 phyically seperate radios both transmiting to 2 radios in box1), I was never able to achieve full speeds, when both were active at the same time.
So:
both radios in same box with antennas close - massive speed decrese (50-90%)
both radios in same box with antennas seperated - speed drops (35-50%)
Even using japans frequencies (ie. one radio on 2412 and the other on 2484) there were definite speed drops.
Other things I noticed:
When one radio is on 5mhz channels and the other on 20mhz channels (the 5mhz channel only takes a 15% hit, the 20mhz channel with take a much larger 50% hit)
If you are using 5mhz and 10mhz channels together on 2 seperate radios they do not play nice 5mhz channel takes very large >75% hit, and the 10mhz channel only takes a 10mhz hit.
To battle this one-sided win, I locked the radios down at 24mbs and the hit on both link evens out a bit. If you watch the receiving antennas status, you’ll notice that when alot of links are transmiting that they will jump around alot (ie. 6mbs to 18mbs to 36mbs to 54mbs, back down etc.)
If you reduce the number of available options it can increase performance greatly.

This is rough estimates of the results, I am sorry I cannot provide more detail, I do not have all my ntoes with me. I am greatly looking forward to see if Mikrotik adds an option to disable the CDMA backoff function in v3.

7

I had a lovely time with 500ms pings when it would ping at all. The AP only had one client on it. The signal was great and the noise floor was great too.
Changing channels fixed the problem.
The bitch of it is, is that you can’t see anything of any value in MT to find out what the problem is.
The old Zcom radios showed all kinds of error information, so why can MT?

8

This is very intresting topic.
I’ve had the same issue and at the end the soltion was to set the 2 cards at the same channel :question: while non overlap channel show eretic ping on client on a card or in the other alternativ.
My hardware is a RB532A with 4 miniPCI (CM9), but also I have tried with RB112 with 2 miniPCI (CM9).
I’m asking myself, Intel miniPCI how they works on RB, have someone test it? :unamused:

Ciao

9

We are having same problem, with a PC router with a fully loaded RB14. We were planning on running all 3 radios in 2.4Ghz on channels 1, 6, 13 but this didnt work well at all.

At first I thought it may be because it has a via chipset on the board and it was not sharing resources properly. However the 4th card on the RB14 is running on 5ghz and this link is rock solid. I also thought it might be interference from the PC PSU.

Anyway we have temporally got around it by using 10mhz channel spacing on the one radio.

Can anyone confirm that running 3 cards on a RB14 on same pc on 2.4ghz will give bad results? Are the radios to close to each other on the board, dont they cause massive interference ?

Thanks

10

we’ve seen this in 5GHz Band when using nearby channels. We now
keep a distance of at least 1 channel between and have increased
performance significant. I think the story with non overlapping channels
is an idealized model. There seems to be an effect on nearby channels.
So I try to get maximum distance of frequencies and Board/Antenna
placement.

Stefan

11

Anyone have any new information on these issues, or explanation of how you got around/fixed it?—We seem to be having the same issue currently.

Fixed it by spacing (10-mhz) frequencies much further apart (3-card 2.4ghz AP system) with 2 cards, but can’t turn on a 3rd card due to the issue. 25mhz apart, and the two card seem to be ok now, but not enough spectrum for the 3rd card when hiding in 10mhz portions between major 20mhz channels :frowning:

We have another 5GHZ system tha we cannot get to work no matter what we try, as we only have limited available spectrum to play with. Testing out the 3.0beta4 with CSMA helped slightly but completely shot latency/ping times up and all over the place.

How about splitting up the cards into multiple systems and placing then 2-3 feet apart? Would that help, or would the cards still be too close to each other still ? How about moving antenna further apart, or both? What did everone else having this issue end up doing?

Good thing I never lit up our bonded backhaul link yet, as I can only imagine the problem that would have had…



SMA

12

I have about 30 WDS APs running with SR5 and CM9 radios.
I’m using them as an ad hoc mesh, almost all of them connected directly to one base station where I have 3 ADSL connections and one dedicated DSL “very expensive” connection.
It all works fine for minutes, and suddenly bandwidth drops abruptly, almost as if they were all disconnected. Stays like that a few seconds, and we’re up and running again…

There are two central RB532 each with two SR5 radios. Two @5700GHz and the other two @5500GHz. I don’t use more than 2 Mbps (when I can get them), and normally everything works fine. CPUs work almost all of the time at 5% of usage, but when bandwidth drops, CPUs go to 100% usage, stay there, and back to normality.
Of course, I assume CPU usage is dropping my network, but,

COULD CDMA BACKOFF FUNCTION BE CAUSNG THIS?

I studied the Log, but I see nothing strange there (at least, nothing that happens exactly when CPUs are heavily loaded)

Where should I look? How can I test this? :question:

Bandwidth (still) falling
13

Chofex, check signal quality.

14

We have completed further testing on SR5 5GHZ cards today, related to this issue.

Formerly we had 4 SR5 cards in one system, at 10mhz of spectrum, in adjacent channels (5745,5755,5765,5775). As mentioned in my prior post in this thread, that caused major loss of thoughtput, loss of connectivity, etc…
When tested later on, we founf 40mhz of spectrum was required to space these apart—and things operated properly at 5745, 5785, 5825—leaving on 3 cards able to be operated in one machine in the 5.8ghz spectrum.

Today we split up the card into two different machines, rackmount metal cases located indoors, seperated about 2 feet apart from each other, in the same small 6’x5’ room. One system had card set for 5745 and 5785, and the other at 5765 and 5825—all set for 10mhz frequencies. That resulted in a minimum of 40mhz of space between card in either individual machine, and atleast 20mhz between each machine. Testing showed no loss of thoughtput or packet loss whatsoever this time, and all 4 cards are useable now. There was issues with adjacent 10mhz channels even on different machines however, so 20mhz is really needed there, and still 40mhz spacing insie the same machine. This gives you a maximum of 3 cards in one machine, or 5 cards across 2 machines (2 + 3)–based on required frequency spacing.

Your situation may be slightly different, if you can spread the machiens much further apart, or worse if you use regular tower cases (and not heavy duty thicker rackmounts as we did). Using outdoor mounted APs may yield completely different results for you. This did not take into case antennas either—we used (4) 90-degree vertical sectors. Mixing and maching horizontal and vertical sectors may have yielded slightly different results.


We will be continuing our testing on the 2.4GHZ spectrum, in 10mhz frequencies this week, and will report back on results. We are configuring every thing to rise in between the 1,6,11 major 20mhz channels, so that leaves very little room for us to play in. As it is, in one machine, we could only get two SR2 2.4GHZ cards to function well in one machine, utilizing 2422(or)2427 and 2447(or)2452. Not sure how we are going to get more than that when using two machines, but we’ll see what we can come up with :slight_smile:

Hope that helps some of you guys out…

SMA

15

Also, we contacted Ubiquitti directly about the 10mhz spectrum usage, and they stated quite clearly:

<<< The AR5006 Atheros chipset does have internal RX filtering scaled
down to 5MHz. We have AR5006 products out today (LS2,LS5, SR4,SR4C) and others to come in Q1.

The AR5004 (SR2,SR5,SR9) based Atheros chipset does have internal RX
filtering scaled down to 10MHz (despite the misconception of 20MHz
only). Interestingly, in our testing AR5004 performs better than the the new Atheros chipsets. >>>

So, contrary to popular belief around here, Ubiquitti is stating that the SR2/SR5 DO support both RX & TX at 10mhz of spectrum.


SMA

16

maxfava, All signals are near -80, but yes, I do have 2 or 3 clients that connect and disconnect from time to time (say -87 or -90 dBm)
Is this affecting performance?
Should I ask for a minimum of how much?

17

Chofex,
for the delay,
in teory if you use ack indoor a client with low signal does not create problem to the other. if you set ack to dinamic it does.
but in practice when a client has a low signal it create problem to the whole AP since the AP try to resend the lost information to the clients, and this create overhead.
also try to switch in B-mode only and change the channel.

Best Regards
Massimo

18

This is really interesting stuff, thanks for the research :slight_smile:

Stuff like this would be great sticky or wiki

:slight_smile:

19

When all my networks will be stable and I can eat a Pizza with my friends :laughing: then I will have time to spend to the wiki. :wink:

20

Yes, I am the same.

V3 should help with this I hope