Page 1 of 1

Low Throughput - Why doesn't bandwidth get through?

Posted: Thu Jun 18, 2009 8:41 pm
by Chofex
We have a more than 200 clients WISP, that distributes 10 MBps of SDSL through a HotSpot to a 5.8 Mikrotik network of 20 nodes, each retransmitting in 2.4 to clients (all clients same model and brand)
We can't get good throughput among each client and repeater. If we measure a client individually, it seems reasonable bandwidth; but adding all the bandwidths at a repeater, they sum up very little.

And it's very strange, because if we measure bandwidth from a client to the repeater, we get 1 Mbps, from repeater to central node, 12 Mbps, but being 40 clients connected to a repeater, each one trying to get 256Kbps bandwidth, they get less than 500Kbps as a whole!

We're pretty sure that trouble is at repeaters. They are RB433AH with an R52 connected to a grid and receiving from main node at 5.8, bridged with rstp to an XR2 with omni. RSTP Bridge has priority and admin mac well defined in each node.
RouterOS is 3.25

we've tried changing in a node to all clients mikrotik, but it didn't change a bit.

We use 2.4-B at different frequencies in each node.

Problem gets worst at peak hours (20:00 to 23:30).

What would you try?

Re: Low Throughput - Why doesn't bandwidth get through?

Posted: Sat Jun 20, 2009 6:29 pm
by Chofex
Ok, seems that I'm not making an interesting question... shame of me!

Despite of that, I DO have this problem, and my customers aren't very fond of it...

What COULD I do?


Re: Low Throughput - Why doesn't bandwidth get through?

Posted: Sun Jun 21, 2009 7:58 am
by jonesy
A few clients modulating at low rates will kill performance for everyone else on the AP, check your registration tables to see what modulation your clients are connected at. Also, lots of small packets (eg. VOIP rtp traffic) will degrade performance.

Re: Low Throughput - Why doesn't bandwidth get through?

Posted: Sun Jun 21, 2009 4:04 pm
by MyThoughts
This is in part a problem with the way 802.11 was designed.
It was not designed with having large numbers of simultaneous connections.

You have an AP (5 GHz or 2.4 GHz, doesn't matter) you test it with one client and get 12 Mbps
Conventional wisdom would tell you that you could have 12 x 1 Mbps connections and be fine, this is NOT true.
I have not really delved into the 802.11 design too much, but I theorize that the reaons for the drop in aggrogated speed has to do with:

CSMA - at both customer and AP, the more radios that are attempting to Tx the more likely you are to have backoff
Retransmits - one link on the entire AP can cause huge performance losses if it is a bad link, the more clients the greater the chance
HDX - most customer presence equipment is HDX (since AP can't Tx and Rx simultaneously the more customers the more switching which takes time)
Design - staying connected to the AP required Tx from both AP and CPE, the more client the more 'wasted' transmissions

If I were to lay blame on any one thing it is the CSMA issue. I harp on this all the time, but I have tested extensively and found it to be very common. For anyone that doubts me run this simple test.

AP1 (2ghz-5mhz/2412, or 5ghz-5mhz/5800) -> CPE1
AP2 (2ghz-5mhz/2432, or 5ghz-5mhz/5820) -> CPE2

You can use 5 GHz or 2.4 GHz, all that really matter in the end is the strength of the signal at the opposite AP (if you use 5 MHz channel sizes the affect is even more pronounced), seperate the channels by a good 20 MHz. In theroy the links should be able to individually operate and max speed, but run a bandwidth-test Tx-only from AP to CPE. Test them individually (other link shut down), indivudally (other link operational but no data transmission), then simutanesouly.

Last time I did this you could acieve ~90% max bandwidth of each link when operating simutanesouly only when seperation was >30 MHz, any less and the percentages drops. Using alternate polarizations helps a bit, seperation helps, but any two tx antennas operating in 'listening' range will cause this.

I've completed this test in a lab enviroment and on 3 towers simultaneously to see how they were affecting eachother the results were not to my liking. I have not had the luxury of completing these tests with cavity filters to essestially negate the signals are much as possilbe before entering the radio, if anyone completes a test with those please post your results.

If you take a look at the protocols used in WiMax, in cell phone systems, and in the RouterOS competitions products, they attempt to account for these problems is various ways.

The problem is made worse do to the popularization of VoIP, online gaming, streaming video, etc. anything that cause as the previous poster mentioned continously small-packet transmission.

I would love to see a RouterOS expand Nstreme to allow two radio cards to function as one in the AP, ie. one card Rx only, the other Tx only BUT allow CPE's to connect with only one card, this way the AP is FDX but clients are still HDX. Not sure is the current 802.11 radio technology could be expanded to do this or not, but it would be a big improvement in efficiency.

I would also like to see RouterOS be expanded to allow AP sync, this allows better utilization of the very limited spectrum most of us have to work with.


Re: Low Throughput - Why doesn't bandwidth get through?

Posted: Sun Jun 21, 2009 5:17 pm
by Nuke
I agree with MyThoughts. Also using an omni is the worst you can do, more so in high noise areas. By swapping to 3x 120deg sector's I believe you will have alot less issues. I tested that with an omni I get far lower speeds than with a good quality sector. Less than 1mbps on the omni with 5 clients compared to almost 4mbps with a sector with 15+ clients connected.

Re: Low Throughput - Why doesn't bandwidth get through?

Posted: Mon Jun 22, 2009 5:14 am
by Chofex
OK, thanks a lot for your comments!

Let me tell you a few things we've tried.
First, we wanted to know if clients with bad signal would drop overall perfomance of the AP.
This turned to be false. Low signal on clients doesn't change the performance of other clients with good signal.
We made a test bench for this, made an AP with 5 clients connected, 3 of them with VERY low signal, one was permanently connecting and disassoc, yet, bandwidth to a well connected client wasn't affected.

Secondly, we place a second AP very next to the first AP, put a lot of traffic in it. Yet, performance on the client connected to AP1 wasn't very affected.

Third, I checked rates of 17 clients connected to one of our APs. Only one had 1 Mbps, all of the others were connecting at least at 11 Mbps. So I started Mikrotik BandWidth test on my PC, connecting to a central RB433 which the AP connects to (so AP's CPU is free), and started test. It showed an erratic bandwidth anywhere between 50Kbps and 1.2Mbps. Y removed authentication to the client rating at 1Mbps, so now all cilents were at least 11Mbps. Nothing changed at the bandwidth test I was running. It continued to flow randomly. Allowed authentication again... nothing changed.

Thing is I used to have more bandwidth when connected to this AP. More than 1.5Mbps and quite stable. But now, with no apparent reason, I seldomly measure more than 200Kbps... and I have plenty free bandwidth to take from at my central node!

We are studding the possibility of installing a notch filter... something like this:

Please, DO feel free to post your experiencies.
So far, thank you very much for your posts! (specially yours MyThoughts, you really took time to explain yourself, I'm glad you did it)


Re: Low Throughput - Why doesn't bandwidth get through?

Posted: Mon Jun 22, 2009 5:20 am
by Chofex
Nuke, in our experience nothing changes if we replace an omni with 3 pannels...
We've done the experience. Signal improves a little, but there's no significant bandwidth improvement at all.
We are placing many nodes, so distance to clients is quite near. That way, omni antennas work pretty well.

I will double check on this. It's been quite a long time since we did benchmarks trying various antennas. If I find significant improvements, I'll post inmediately.

Thanks anyway for your post! I'm sure anyone needing better signal will find it handy.


Re: Low Throughput - Why doesn't bandwidth get through?

Posted: Tue Jun 23, 2009 6:40 am
by MyThoughts
I run all my site with the newer modulation (OFDM), so 802.11g,a,5mhz,10mhz the affect of CSMA on 802.11b is much less as it doesn't pack as much throughput per mhz of spectrum. I have never run these tests with 802.11b.

Could you post some of your bandwidth screenshots and data?


Re: Low Throughput - Why doesn't bandwidth get through?

Posted: Tue Jun 23, 2009 4:37 pm
by Chofex
Here you can see one of my nodes, it receives signal using 5.8 from our NOC, and transmits to CPE using 2.4
18 CPEs can't use that little bandwidth!
If I run a Mikrotik bandwidth test from a PC connected to a CPE connected to this node, running BWTServer at my NOC (connected by 5.8 to this node), I can get this morning TCP 4 Mbps down, 1.5 Mbps up, and I see that traffic at node's wlan5.8.
I did it last night, while everyone was surfing, and I got less than 0.5 Mbps, and wlan5.8 never got more than 500Kbps.
It's like if everyone is trying to get bandwidth, then noone gets it!