Low NV2 802.11ac PTMP Thoughput

gmsmstr · Thu May 07, 2015 11:54 pm

Have an access point, netmetal running dual chains to 120 degree sector.

/interface wireless
set [ find default-name=wlan1 ] adaptive-noise-immunity=ap-and-client-mode band=5ghz-onlyn channel-width=20/40mhz-Ce default-forwarding=no disabled=no frequency=5265 ht-basic-mcs=\
    mcs-0,mcs-1,mcs-2,mcs-3,mcs-8,mcs-9,mcs-10,mcs-11 ht-supported-mcs=mcs-0,mcs-1,mcs-2,mcs-3,mcs-5,mcs-8,mcs-9,mcs-10,mcs-11 hw-retries=15 l2mtu=1600 mode=ap-bridge nv2-cell-radius=20 nv2-preshared-key=key \
    nv2-security=enabled radio-name=TD-120-East rx-chains=0,1 ssid=east tdma-period-size=4 tx-chains=0,1 wireless-protocol=nv2-nstreme

ss68 May. 07 15.51.jpg

As you can see 45 clients.

When we started with about 10 clients, we could get 50-60 meg per client, in ac mode.

Now with this, in a/n/ac mode we get less than 5 meg to each client, in N only mode we get around 10ish at most. we never see the access point go above 30 meg transmit!

All clients are SXTs, everyone is mostly above -70 .. clients with -60s are only able to get 6-7 meg if that.. in the evening is slows down even more with the ap only dishing out around 10-15 meg total.

Any help here? we turn on nstream, and get polling timeouts, and in 802.11 mode we barely can even winbox to clients

Lupin · Sat May 09, 2015 10:11 pm

You should try to make BWTest TCP client with the best signal.
In meantime you manual disconnect other users until you find the one (or ones) that degrade performance.

With Mikrotik often I found that clients, because obstacles, degrade the entire cell (even if the signal is good). They should work a little more on the protocol PtMP

In your case I would check
E7: 7D
76:15
72: C1

And remember that if there is an obstacle between a client and your AP mounting an high gain antenna, doesn't solve the problem, improves the situation of course, but do not solve

I think Mikrotik can work even on two things:
1 - prevent the degradation of the cell if only one client has problem
2 - adjust the protocol (PtMP) to work even if there are simple obstacles (branches of a tree). Today if I have a -60 (which is a good signal) with 2km distance with a QRT5 the customer goes very wrong. If I have a -60dB at 6km, QRT (for example) going very well

pacoss · Sun May 10, 2015 12:43 pm

Remove clientes over 70.

That CCQ's are worst than bad. You're not working in a clean frecuency.
Be carefull when scanning with the Ubiquitis sectors shifted +5Mhz, you can't see them but they are there. I have a cheap Omnitik installed in nearly every places (with wlan1 disabled) only for scanning purposes.

Give a try using 20 Mhz channel width in sectors, it's easy, fast and safe. Force some traffic to all the clientes, using ipscan tool for example, and then look signals and specially CCQ. Then, test some of the clients that you have previously tested and compare.

If better ... start to change the AP frequency until you get the best quality and throughput in average.

Regards.

WirelessRudy · Sun May 10, 2015 1:27 pm

OmniTik as extra unit in AP location is a very good Idea I am using since some time.
Some advantages;

1. If the PoE-out version is used you can power control 4 units remotely (or automated by script). Like 1x backhaul, 2-3 AP sectors.
2. You can always perform a radio scan or Sniff. It won't disturb the actuall working of the real AP's.
3. If you'd use any local traffic processing like mangle, filter, QoS, you can have it done by the Omnitik. Simplyfies the setup of the real AP's and leave the real AP's cpu free for traffic handling only.

If running a wireless scan or sniffer (sniffer 'sees' frequency usage not noticed by scan) make sure to do also a 10 and even 5Mhz scan.
I found an operator using several 10Mhz centre frequencies the same as my 20Mhz ones.... (not very nice!)

Last, but not least:
Sometimes it can be it all looks fine at the AP. No close, or very weak, neigboring frequency usage scanned.... but; it can still be that at the client its different!

We had one AP with only 8 clients, ranging 15meter to 300meter max in a town.
All clients had signals ranging from -50 to -35. But still we had one client with - 37 that had a very poor CCQ (5-20%) We had problems to bring more than 2Mb to the client.. but the scan on the AP didn't reveal too much. Only one 10Mhz channel usage of competition 20Mhz beside our centre frequency and signal strenght measured at (at the Omnitik with is 7-8dBi antenna!) -88 - -93dB.

When we did a scan on the client we found that same providers AP hitting our client with -30dB! Stronger than our signal!
We swapped the AP frequency completely away and the problem dissapeared!

How to scan remote client in 10Mhz band while it should be connected to 20Mhz working AP?

- Set 10Mhz channels in the client's 'channel list'.
- Set in the scan part of the wireless both the AP's centre frequencies and 2 or 3 adjacent 20Mhz channels.
- Set in the scan part of the wireless also all 10Mhz frequencies from the AP's frequencie and into 25Mhz to both ends.

Basically you have now setup a 50Mhz wide band for 10Mhz scanning, and a 100Mhz wide band for 20Mhz.

Now, log into the AP, open a telnet session to the client and run the scan;

int wi scan 0 d=10

The client will now scan for 10 secs in the channels given in the scan list of client.
Client will disconnect but when it comes back, and after some time (10-30 secs) the telnet session in AP comes alive again and will show the result of the scan!

Now you can see what is actually happening around the working frequency of the AP, but AT the client!

Good luck!

WirelessRudy · Fri Jul 17, 2015 2:01 am

Lately sort of discovered the spectral scan tool. I knew it existed but never really took a good look into it, until I found that a normal scan doesn't see a lot and you need to set the scan frequencies in the 'scan' portion of the unit you need the scan off.

What I do now is do a remote (so from the AP's winbox session I open a telnet session to the client) frequency-history. You set the range and some time frame. Let the scan run for instance 30 secs after about another 20secs or so the remote unit returns the results in the telnet winbod. (The telnet session doesn't break, like a winbox session does. A telnet session just waits until unit comes backup, even after 5 mins....)

So, find the IP of the CPE and open under / tools a telnet session (there are other ways to open one but for simplicity this one)

after typing your admin and password the remote unit opens in telnet window (that runs in the AP winbox session)

type

int wi

<enter> the unit opens in the wireless interface menu section,

pr

<enter> and the unit returns with the wireless interfaces and its setting,

spectral-history 0 range=5100-5950 d=30

<enter>

Now you'll see the CPE disconnect after some time but thats normal. Wait until the CPE connects again and wait about half a minute more (depends on how good the connection to CPE is and how many other CPE's connect to AP)

Now, after this time the still open telnet window (before typing the codes to do the scan widen it to the whole screen, that gives best resolution) shows the spectral scan result in color! Very nice. You probably see a yellow or red color band where your AP is transmitting and all other signals around. Now you can see if your spectrum at the client is really used and if there is anything free you might think to move...

To guarantee that another 3rd party source is not interfering with your AP's working frequency you can switch your AP's radio off for about 20 secs just after the start of the CPE's scan.
In the graph you can now see your AP's spectral usage drop off and come back. If it still shows yellow colors close to your AP's working freq. even when it was off, you know some other frequency is disturbing yours....

The advantage of this spectral scan is that the scan setting in the wireless part of the radio at the CPE can be left as it was. I usually only set the frequency of the AP so the CPE comes back asap after a disconnect. It doesn't have to 'search' in the band for the AP.

Second advantage is that the spectral scan will pickup 'any' frequency (802.11? I actually don't know about non 802.11 radio sources) in the band. 5Ghz, 10, 20 or 40 and with or without shift.

You can do the same in the AP or other CPE's to see if other places are bothered by the same interference sources.

In the AP I also do a scan with 10 and 20Mhz wide channel. It gives me the names of the other networks so I can identify which network is the one that disturbs mine! (I found to my shame at some occasions it was actually one of my own! But ok, I use 50 different radio's!)

Now, one reason why an omnitik is a very good idea to have at your AP: MT 802.11ac can't do the spectral scan!
So, if all your CPE's are also ac than your thrown back to the use of the 'scan' tool in the winbox. That's a lot more complicated to do and doesn't give all the info you want.....

I really don't understand why bringing back the spectral scan tool in ac units again. But I've been told its more a shortcoming of the chipset than the software...
But for me its one reason why I still won't use any ac CPE units. What is the use of ac in PtMP networks if I hardly can 'see' what the spectral situation will be.
ac is partially a solution for spectral issues, but you need a tool to at least use it than to the advantage. Now its not....

Lupin · Fri Jul 17, 2015 6:29 pm

The basic problem is that when you connect many customers on a cell the bandwidth decreases drastically.
I have the same problem, about 45 clients connected with optimal signals, but I can not pass over in 6-7Mbps TCP in download (test on the customer with best signal)
The access poin at that time also gets little traffic (10-12Mbps) and with my test coming to a total load of up to 20Mbps.
The absurd thing is that if I leave only this customer attached on the cell, it arrives at 40-50Mbps TCP.

I am increasingly convinced that the NV2 protocol in PTMP mode has not been optimized.
It seems that when a CPE is not in optical visibility (although with good signal) this goes to heavily degrade the cell.

There isn't interference on that frequency

InoX · Fri Jul 17, 2015 8:26 pm

As I stated many times on this forum, more than 20 stations on a mikrotik AP will result in very low bandiwth, high ping and also serving timed outs to clients. NV2 is not good, nstreme the same.
With nv2 and 15 stations have ~15ms average ping at ~30-50mbps trafic.
I can't imagine 45 stations...

Lupin · Fri Jul 17, 2015 9:46 pm

If 40 customers share 60-70Mbps overall bandwidth then OK, it's saturation.
But, if 40 customers are not making traffic and, just for the fact of being associated with the AP, degrade the performance of all, mean that there is a problem to resolve, especially if a station reach max of 8-10Mbps when on AP there are only other 2Mbps

For example I have a BTS with 8 Mikrotik cells, each charge from 25 to 45 customers and I really see how it is behaving.
I can not add more cells for lack of frequencies.
I'm realizing that there is no solution with Mikrotik.

I asked many times to improve performance in wireless technology for WISP, but they doesn't seem to have an answer.

I saw other providers use the Cambium Network PMP450 and connect 100-120 customers on a 20Mhz cell and I see customer on that, reaching bands of 40-50Mbps with incredibly low latency

There are ways to improve and I think the Mikrotik team knows them, just not seem to develop.

Unfortunately, I have areas where I do not have ability to create other BTS and new customers continuously to arrive, but I have no possibility of growth.

It would be good news if Mikrotik announced that they are working on something on the wireless world for Wisp (eg GPS Sync) so I can stay with them, but this is only a hope...

WirelessRudy · Fri Jul 17, 2015 11:27 pm

Can't agree more with you guys.
In the 'old' days, when tdma was grown out of its initial problems it worked very good for me with some AP's up to 40 clients, single chain even so!
But now with the tough competition from more operators to serve more clients that on average also are more demanding we are running in the same 'trap' as mentioned here. Too many customers with too many bandwidth assigned to them in a too crowded spectrum.

Did we offer 'up to 6Mb' speed for top contracts 2 years ago, that is now considered basic.
If clients didn't get their 'top' speed regularly two years ago, hardly any complained. For an occasional download of a movie or photos or some occasional browsing not such a big deal. The network performed relatively well too so not a lot of complaints. Nowadays it looks like some make it their habit to open their PC with www.speedtest.com and if its not top, I'll get the message....

At the same time came the many, many (50 -80 up to 200Mbps) deals from the big carriers in the big towns, and together with the push to buy the smart TV with the embedded streaming video and TV options that people wanted to use, even if they went to live in the country side, people expected that kind of internet to be available everywhere.

"I bought this internet smart TV but it doesn't work".... "Well, you have a basic contract with a poor connection that can only do your assigned 2-4 megs...", ... "Well, that ain't good enough, the shop told me my 50" 3D HDTV should also work in the country side....."

And oh yeah, now he complained anyway, amidst the crisis; "How can it be the competition offers double the speed for half the money?" "Maybe I should go swap operator??"

Every customer we loose needs two new ones just to keep up with the income....
And more new players entered the market, with new lower priced contracts and new towers working in the same band molesting my towers.....

Same time this country (Spain) was years behind compared to the North of Europe, special in the country side, but that is now being corrected very fast. So still we are getting new clients every day, once we need just to get the money in to upgrade as good as possible our network...

So here is the problem. Existing networks have to deliver more bandwidth in worse spectrum to more clients for lower income per customer.
I found out indeed 25-30 per AP is going to be an absolute max. and only the best antenna's are good enough to get as much as possible out of the spectrum.
With Mikrotik it is becoming a very difficult task and as I am being told with ubnt its not so much the better.
Some on this forum actually also stated that during a MUM he spoke with one of the MT guys and it was actually confirmed that wireless was not really one of their focusses...... so that makes you to start thinking.....

I already looked at Cambium, looks very promising, or what to think of Mimoso. Looks even better.
But where the hell to find the money and the time to change my network.....

HaQs · Tue Jul 21, 2015 10:56 pm

I confirm this some problem. (on noisy area, and more >15 clients on ap)

and on the new controller "wireless-fp" Pings little better on NV2, but the stability is worse.

In the old driver more kept connection.
On the new lot more disconnects in the noisy area.

I hope that implementation can improve MU-MIMO case.
If not, unfortunately, we will need to migrate to another manufacturer.

Lupin · Tue Jul 21, 2015 11:54 pm

I'm already considering another supplier.
I think that Mikrotik never improve its system for wisp

I begin to have many customers and situations of degradation are more and more.
Are ONLY two years in which I hope some improvements:
(http://forum.mikrotik.com/viewtopic.php?t=76376) Sep 04, 2013

I will look MUM Friday what they show, but will be the usual products

Pity, for € 15,000 (sxt, sextant, qrt, netbox, netmetal) of material that I order every month, go into other pockets

gius64 · Fri Jan 29, 2016 3:25 am

It's very disappointing to see that there isn't a single word from a MikroTik guy about that

I invested lots of money in MikroTik for my network and I don't see nothing they do to tell us how to improve performance in PtMP.
If they think performances are ok, they could write a wiki article about how to set up a good performing sector with more than 25-30 users on it.
Competitors performs better than MikroTik with the same spectrum and more users.
Cambium states they can reach up to 120 customers on one sector. On MikroTik that's impossible!

I hope they have a solution for that. They can't leave all WISP alone without solutions to grow.

I wouldn't like to go with Cambium, because it would mean change all my existing SXT and so on... And I like MikroTik system, but they have to improve wireless a lot!

They released some sectors, maybe they are preparing something new for PtMP?
Maybe a new protocol in RouterOS 7 which will help?

MikroTik, please, give us some feedback!

server8 · Fri Jan 29, 2016 11:57 am

We have APs 20Mhz with 50+ clients. Signals are under -50 with only some clients over -60. The AP has a maximum aggregate of 50 Mb/s in dramatic high noisy enviroment.

Client side bandwith test reach 50 Mb/s with signal around -40 and only 7 Mb/s with signal around -65. Generally we use shielded SXT AC 90° as AP and high gain antenna client side.

We apply queues client side: with good signals (under -50) we give 10/1 Mb/s with bad signals (over -65) 3 Mb/s/512kb/s. In this way clients with bad signals don't takes airtime to retrasmit loss packets they have slow connection.

I hope to see soon NV3 and sink.

Giuseppe

gius64 · Fri Jan 29, 2016 1:04 pm

We have APs 20Mhz with 50+ clients. Signals are under -50 with only some clients over -60. The AP has a maximum aggregate of 50 Mb/s in dramatic high noisy enviroment.

Client side bandwith test reach 50 Mb/s with signal around -40 and only 7 Mb/s with signal around -65. Generally we use shielded SXT AC 90° as AP and high gain antenna client side.

We apply queues client side: with good signals (under -50) we give 10/1 Mb/s with bad signals (over -65) 3 Mb/s/512kb/s. In this way clients with bad signals don't takes airtime to retrasmit loss packets they have slow connection.

I hope to see soon NV3 and sink.

Giuseppe

How did you configured queue client side? Did you find it faster than AP side?
You could post your configuration here if you want: http://forum.mikrotik.com/viewtopic.php?f=7&t=104285

imagica · Sat Jan 30, 2016 10:41 pm

Well i had been facing exactly same issues with Mikrotik NV2,performance degrades after 25-30 clients,latency keeps between 15-20 ms.

I have made following changes to my network and now everything has returned to normal. You can try these changes and check if it helps you.
1) all client's are kept in between -50 to -60 range
2) upgraded all sector anteenas to ubnt ac 60 degree 21 dbi powered by basebox 5
3) changed ap settings to nstreme as discussed in this post
http://forum.mikrotik.com/viewtopic.php?f=7&t=83183

Surprisingly nstreme never worked for me with default settings but with this modified setting it is working amazingly fast with no disconnections.Version currently being used is 6.33.5

Also i run my ap's on 10 mhz channel and max bandwidth i provide is 10/10 Mbps.

Currently have deployed this over 45 sectors and 10 PTP running combined bw above 500 Mbps during peak hours.

gius64 · Sun Jan 31, 2016 6:03 am

Well i had been facing exactly same issues with Mikrotik NV2,performance degrades after 25-30 clients,latency keeps between 15-20 ms.

I have made following changes to my network and now everything has returned to normal. You can try these changes and check if it helps you.
1) all client's are kept in between -50 to -60 range
2) upgraded all sector anteenas to ubnt ac 60 degree 21 dbi powered by basebox 5
3) changed ap settings to nstreme as discussed in this post
http://forum.mikrotik.com/viewtopic.php?f=7&t=83183

Surprisingly nstreme never worked for me with default settings but with this modified setting it is working amazingly fast with no disconnections.Version currently being used is 6.33.5

Also i run my ap's on 10 mhz channel and max bandwidth i provide is 10/10 Mbps.

Currently have deployed this over 45 sectors and 10 PTP running combined bw above 500 Mbps during peak hours.

How much users can you put on a sector with this configuration?
What about limiting users? Do you have default queue types?

imagica · Sun Jan 31, 2016 11:45 am

Since i have been testing this new config for last 10 days currently all sectors are having subs in between 25 - 35. Am planning to add more and test the result. Till now everything is working smoothly.

ronniee · Tue Feb 02, 2016 9:06 am

What about limiting users? Do you have default queue types?

Can you tell us how do you make limitation to users? Simple-queue on AP? PPPoE?

p3rad0x · Tue Feb 02, 2016 11:44 am

I have a sector with 45 Clients also used to perform like garbage no matter what frequency.

Two days back we had a lightning storm and I think it killed one of the other providers neighboring tower.

Client with the worst signal -77/-72 now gets 13Mbps down and 2 up Client with the best signal -51/-47 now gets 35Mbps down and about 21Mbps up.

This is on 20Mhz wirless a/n with a HT MCS 7.

I'm expecting the performance to become cancer once the other providers tower gets fixed.

They do 40Mhz full power 31dbm, complain to the Communications Authority and they do nothing about the problem that a unlicensed wisp is taking up all the spectrum.

server8 · Tue Feb 02, 2016 3:28 pm

How did you configured queue client side? Did you find it faster than AP side?
You could post your configuration here if you want: http://forum.mikrotik.com/viewtopic.php?f=7&t=104285

We use client side queue like this

/queue tree
add max-limit=10M name=ether1-root parent=ether1
add max-limit=512k name=wlan2-root parent=wlan2

In this way all AP cpu power is used for NV2.

Giuseppe

gius64 · Tue Feb 02, 2016 3:35 pm

How did you configured queue client side? Did you find it faster than AP side?
You could post your configuration here if you want: http://forum.mikrotik.com/viewtopic.php?f=7&t=104285
We use client side queue like this

/queue tree
add max-limit=10M name=ether1-root parent=ether1
add max-limit=512k name=wlan2-root parent=wlan2

In this way all AP cpu power is used for NV2.

Giuseppe

Good idea, but you can't have a centralized limit management

Did you find it's faster than having limits on AP side?

server8 · Tue Feb 02, 2016 4:51 pm

Good idea, but you can't have a centralized limit management
Did you find it's faster than having limits on AP side?

We have a script running client side every night that set the speed checking the wireless signal, if the signal is good the client has the max speed if the signal is bad we set lower speeds. In this way clients with bad signals take less airtime in retrasmitting lost packets. They will have slow internet connections.

imagica · Tue Mar 22, 2016 7:46 pm

What about limiting users? Do you have default queue types?
Can you tell us how do you make limitation to users? Simple-queue on AP? PPPoE?

We use CCR-1036 as B-RAS for bandwidth limiting alongwith radius server for billing and authentication.

WirelessRudy · Mon May 16, 2016 8:56 pm

Well i had been facing exactly same issues with Mikrotik NV2,performance degrades after 25-30 clients,latency keeps between 15-20 ms.

I have made following changes to my network and now everything has returned to normal. You can try these changes and check if it helps you.
1) all client's are kept in between -50 to -60 range
2) upgraded all sector anteenas to ubnt ac 60 degree 21 dbi powered by basebox 5
3) changed ap settings to nstreme as discussed in this post
http://forum.mikrotik.com/viewtopic.php?f=7&t=83183

Surprisingly nstreme never worked for me with default settings but with this modified setting it is working amazingly fast with no disconnections.Version currently being used is 6.33.5

Also i run my ap's on 10 mhz channel and max bandwidth i provide is 10/10 Mbps.

Currently have deployed this over 45 sectors and 10 PTP running combined bw above 500 Mbps during peak hours.

So 45 x 10Mhz channels means 450mhz of spectrum usage plus 10 x 20 (at least) makes another 200mhz of spectrum. Thus you are using 650mhz of spectrum in a band that is mostly only 400mhz wide.
So either your antennas are all widely apart from each other and you use high shield dishes and/or you must have interferences issues. Many in this forum state interferences are a killer for nstreme so how do you do it?

The high level of interferences combined with some historic disappoints keep me from using nstreme. But maybe you can make me think otherwise?

The use of 10Mhz channels versus 20Mhz.; I agree that the higher signal strenghts and s/n ratio (= important for mcs rates) of 10Mhz versus the 20Mhz is important, but at the same time, speeds in the network are halved.
The same data to each / a client is transmitted by the radios in a 20Mhz functioning network double as fast as in a 10Mhz network. So radio is freed faster to serve another client. (The only other side of the coin is that high s/n ratio due the 10Mhz bandwidth may allow higher mcs rate. But that usuall doesn't give you double the capacity.)

We skipped 10Mhz a year or so ago since we couldn't deliver the speeds to clients anymore. In same network but 20Mhz we don't have the issue of supplying 2 or 3 clients at the same time with 15mbps. (And apart from that, mikrotik 'ac' radio doesn't support 10Mhz....)

QoS on the client:

/queue tree
add max-limit=10M name=ether1-root parent=ether1
add max-limit=512k name=wlan2-root parent=wlan2

What speed did you set for the parents? Imho if parents have no speed set the child queues don't even work.....
And what would be the use of the 10M limit for the ethernet anyway? If this is the port connected to you client any flooding or saturation on the network already took place before this queue starts to work....
tcp traffic would indeed be slowed down a bit by this queue but actually that should already have taken place in the gateway router where the client limiting took place.
And if it was udp traffic (that was not yet limited in the simple queue of the client in the gateway already) than any flooding of the wireless already took place before this queue could do anything about it?

Queues only work on traffic 'leaving' the router. And since your Ethernet connection is probably of much higher capacity than the wireless anyway (unless you're still on 10M ethernet) this queue has no use whatsoever.

In this way all AP cpu power is used for NV2.

? The CPE just follows what is been told by the AP. Their is not a lot of work involved. Setting up a queue tree imho uses more cpu for nothing than if not set and traffic would leave router anyway.

Imho a proper queue for a CPE should do nothing more than keep the data leaving CPE towards AP (uplink) as lean as possible. So low udp limit and tcp according client's allowed usage. (remote access to his home camera system needs 2-4Mbps upload). In the download (AP => CPE) limiting should have taken place in the gateway (to set client's assigned speeds) and in the AP (or higher up depending on the backhauls) priority given to traffic types over others and maybe per client prioritization.

So although I find you made some steps in the right direction in fact I have doubts on the working in many fronts. And more details are needed to proof your concept if you claim it does indeed work...

Although my memberships states 'guru' that doesn't mean I am knowing all. I am still learning and trying. We also do have capacity issues on the PTMP networks so the purpose of making a discussion about your setup is merely in the hope I can learn something new than to criticize your's.

maniraj4143 · Sun Jun 19, 2016 4:54 pm

I saw other providers use the Cambium Network PMP450 and connect 100-120 customers on a 20Mhz cell and I see customer on that, reaching bands of 40-50Mbps with incredibly low latency

100-120 customer on a 20Mhz cell are really big numbers !
Do you know what cpe are these providers using with the Cambium Network PMP450?

gius64 · Sun Jun 19, 2016 7:28 pm

You have to use their CPE, which costs a lot more than an SXT

maniraj4143 · Mon Jun 20, 2016 3:12 pm

You have to use their CPE, which costs a lot more than an SXT

Thanks for the reply. I'm gonna stick with MT.

shortcircuitonline · Fri Jul 15, 2016 12:28 pm

this example is for ac same appy to 20 mhz or 40 mhz or 80 mhz
if clients or trying to use same time internet .

It strongly depend on wireless network planning. Theoretically max data rate of
mant19s os 866.7 Mbps which in real life is ~600 UDP and less, let's say 400 Mbps
TCP traffic. So in ideal conditions it would be 400/50 = 8 clients per AP. But
let's assume that you will not have ideal conditions then I would say 4 - 6
clients per AP with 50 Mbps in one direction with TCP traffic and 8 - 10 clients
with UDP traffic in one direction.

wireless is not good option in point to multi point
wireless is good only for point to point
i think this way may b i m wrong

WirelessRudy · Fri Jul 15, 2016 1:55 pm

wireless is not good option in point to multi point
wireless is good only for point to point
i think this way may b i m wrong

Your post doesn't make sense a lot... I don't understand. But your last remarks are hilarious. Indeed, we should all stop in what we are doing and deploy cables to our clients.....
Joking aside, Wireless is mostly 'the only' option for P2MP and works good enough. Your cell phone works that way too...

shortcircuitonline · Fri Jul 15, 2016 2:17 pm

ok do test
create a ap and add 1 or 2 client and run mt b test from laptop
and same time do speed test then tell me know.

imagica · Sun Jul 17, 2016 7:43 pm

Well i had been facing exactly same issues with Mikrotik NV2,performance degrades after 25-30 clients,latency keeps between 15-20 ms.

I have made following changes to my network and now everything has returned to normal. You can try these changes and check if it helps you.
1) all client's are kept in between -50 to -60 range
2) upgraded all sector anteenas to ubnt ac 60 degree 21 dbi powered by basebox 5
3) changed ap settings to nstreme as discussed in this post
http://forum.mikrotik.com/viewtopic.php?f=7&t=83183

Surprisingly nstreme never worked for me with default settings but with this modified setting it is working amazingly fast with no disconnections.Version currently being used is 6.33.5

Also i run my ap's on 10 mhz channel and max bandwidth i provide is 10/10 Mbps.

Currently have deployed this over 45 sectors and 10 PTP running combined bw above 500 Mbps during peak hours.
So 45 x 10Mhz channels means 450mhz of spectrum usage plus 10 x 20 (at least) makes another 200mhz of spectrum. Thus you are using 650mhz of spectrum in a band that is mostly only 400mhz wide.
So either your antennas are all widely apart from each other and you use high shield dishes and/or you must have interferences issues. Many in this forum state interferences are a killer for nstreme so how do you do it?

-----
Well 45 sectors and 10 PTP are not on single tower. they are spreaded over 8 towers. Also under Nstreme also if the user count increases above 32 then we have complaints from all over and once we remove some clients and bring it back below 30 then everything returns to normal. So IMHO thats the practical limit we have to cope up. UBNT AC products are performing much better in the same scenario.

WirelessRudy · Mon Jul 18, 2016 4:03 pm

Well i had been facing exactly same issues with Mikrotik NV2,performance degrades after 25-30 clients,latency keeps between 15-20 ms.

I have made following changes to my network and now everything has returned to normal. You can try these changes and check if it helps you.
1) all client's are kept in between -50 to -60 range
2) upgraded all sector anteenas to ubnt ac 60 degree 21 dbi powered by basebox 5
3) changed ap settings to nstreme as discussed in this post
http://forum.mikrotik.com/viewtopic.php?f=7&t=83183

Surprisingly nstreme never worked for me with default settings but with this modified setting it is working amazingly fast with no disconnections.Version currently being used is 6.33.5

Also i run my ap's on 10 mhz channel and max bandwidth i provide is 10/10 Mbps.

Currently have deployed this over 45 sectors and 10 PTP running combined bw above 500 Mbps during peak hours.
So 45 x 10Mhz channels means 450mhz of spectrum usage plus 10 x 20 (at least) makes another 200mhz of spectrum. Thus you are using 650mhz of spectrum in a band that is mostly only 400mhz wide.
So either your antennas are all widely apart from each other and you use high shield dishes and/or you must have interferences issues. Many in this forum state interferences are a killer for nstreme so how do you do it?

-----
Well 45 sectors and 10 PTP are not on single tower. they are spreaded over 8 towers. Also under Nstreme also if the user count increases above 32 then we have complaints from all over and once we remove some clients and bring it back below 30 then everything returns to normal. So IMHO thats the practical limit we have to cope up. UBNT AC products are performing much better in the same scenario.

Well, we have some 40 AP's (Either Omni or sector) with some 650+ clients that have either 6 or 15Mb as their limit but still hardly see more than 250Mb total (download, aggregated we see some 300-325Mb) traffic. And that is basically since we have several AP's with 25+ clients in 20Mhz that still have issues in getting more than 10-15Mb to each client at the same time. These AP's with multiple client bw-test running sort of stop with 50Mbps total traffic where if we do the same test from the AP we can easily get 150-200Mbps per AP.
But apart from these 40 AP's we have some 20 other AP's in the same region from competition and I would guess altogether we have some 35 backhaul links using 5Ghz. Hence I think in our 20km wide region you would probably be able to count 90 radio's all working in the same band as AP. Very 'dirty' spectrum I would say.
We´ve set two AP's now to 40Mhz wide channel were the spectrum was relatively 'clean' and that had 30+ clients to it and that almost doubled the capacity of the AP. We fiddled with nstreme and 802.11 but halve of the clients either lost connection or speeds were absolutely crap. Like I said before, 10Mhz is unworkable for us. Only some AP with only 2 or 4 clients show good results with 10Mhz. But more than lets say about 8-10 clients on a 10Mhz AP is limiting each in its tops speeds....
So I don't know how you do it. We need more info on a particular P2MP network of yours with configs of both a client and AP and it's surrounding spectrum (scan) to see were you've done the magic many of us can't reproduce....

resotat · Wed Jul 20, 2016 7:56 am

Hello All
I offer 10 18 and 22 Mbps packages. I am running RB912UAG5hPnD with Laird 17Db 90 degree sectors and a 20/40 channel width with just a single chain..
I have been fighting this for many weeks now. I hate to rain on anyone's parade but this is not a CCQ issue, nor is it a signal strength issue, nor is it an interference or noise floor issue, this is an NV2 overload issue. The Reality is NV2 can not handle more than 15 connections before it totally degrades the whole AP's performance. I am not sure why this is but I have confirmed it on over a dozen access points.

Here are the numbers with a connection of low to mid 70's it does not matter if the connection is in the 60's or 50's or 40's while obviously the higher signal strength will mean even better speeds you can still achieve the following speeds in Mbps real world connection, at that signal strengths specified, with NV2, if you understand these realities of using NV2.

if you only have one person is connected you can pull 50 - 60 Mbps on a TCP test. between AP and client
If you jump the number of connections to 10, the through put to any given client drops to an average of about 36 Mbps.
If you increase the connections up to 15, the through put drops to an average of about 18 - 26 Mbps.
If you increase the connections to 18, the throughput drops to an average of about 10 - 15 Mbps
If you exceed 18 clients your through put will drop very very very fast you will drop the average to below 10Mbps even taking it down to 2-5 Mbps after 20 - 22 clients connected.

It does not matter what you set the period to. It does not matter if all the clients have a -60Db signal strength (-60 will obviously increase these numbers slightly but overall degradation of the AP will continue to slide on a curve based on the # of clients connected) . it does not matter what channel it is on it only matters how many clients are connected.

Now you must remember this is a bandwidth test from AP to client on TCP not UDP and the real world numbers are about 50 to 60% so If I keep the registered clients to 10 or less I will get at least 22Mbps in a real world speed test.

While Nstream solves the speed issue you end up with weaker connections dropping continuously due to poll timeouts. While having everyone connected with 50 -60 signal would be ideal it is just NOT REALITY and it does not totally solve the poll timeout issue anyways.

Now obviously only having 10 customers to an AP is not practical you will run out of usable channels quicker than you can say &%$# Me.

The big question here is WHEN IS MIKROTIK GOING TO SOLVE THIS ISSUE. Or are we all going to have to switch to another solution to get a TDMA protocol that can handle at least 40 clients and deliver the full capacity of the individual connection to each client.

I have over 1,500 subs and was going to update each sub to a dual chain board probably a QRT as it is the cheapest dual chain solution that is not Ubiquity (sorry hate Ubiquity) but if I can only get 10 clients per AP I will have to spend my $225,000 on a different solution. I really love Mikrotik but this problem makes using it in the modern age of big speed packages unrealistic. I hope they solve this issue in short order.

Wed Jul 20, 2016 11:50 am

@Mikrotik
It would be a good idea to add something to this thread. If there is a scaling problem with nv2 please give a statement and if possible a promise to improve this. If this is a config or installation problem you might showcase an installation which scales and give advice how to install a ptmp system which does not degrade with every additional cpe.
I guess your developers have a test ptmp setup running to verify there code. Please do some benchmarking.

p3rad0x · Wed Jul 20, 2016 2:59 pm

+1

Same issue wit my ptmp connections, starts hitting a magical 12Mbps barrier

WirelessRudy · Thu Jul 21, 2016 1:36 am

+2

I am about to order a new system (not ubnt, hate it!) because indeed we cannot even get close to speeds that other systems are claiming possible. I can't even get the numbers #resotat# is talking about. And that's with duo stream antennas!

I'd love MT for a long time, and complained about their progresses in wireless. But as told them before, they are running behind it just doesn't scale up in the modern world of high speed, low latency GPS sync'd high capacity P2MP networks. It would still be my preferred system in remote areas with not a lot of spectral usage and not too many clients so your cells don't get a lot of customers. But in heavy congested and dense populated areas they need to come up with something new fast. Otherwise their wifi platform is simply bleeding to death...
I hate we are going to another platform, its some butterfly, and we'd probably the first in this country to use it but I see no alternative and we need to go forward.....

WirelessRudy · Thu Jul 21, 2016 1:40 am

I have over 1,500 subs and was going to update each sub to a dual chain board probably a QRT as it is the cheapest dual chain solution that is not Ubiquity (sorry hate Ubiquity) but if I can only get 10 clients per AP I will have to spend my $225,000 on a different solution. I really love Mikrotik but this problem makes using it in the modern age of big speed packages unrealistic. I hope they solve this issue in short order.

If you stick with Mikrotik I can suggest to use the LHG-5 antennas. For CPE they are really good. We have some 40 of these deployed and they are easy to fit, as CPE just as good as an QRT but way much cheaper......
Network tips for any reader;
Backhauls with Netmetals and Dome or shielded directional antennas.
AP's use Netmetals and good Omnis or the RF-element line of antennas. Very good results.
CPE's. SXT or LHG. They are both good and cheap.

vlatko · Fri Jul 22, 2016 11:33 am

we working with mikrotik almost 10 years
im aware about congested spectrum, low ccq clients and simillar problems
they all affect bandwith more or less
but there is magical number on ptmp ap betwen 22-30Mbps
he start to take effect somwhere betwen 5-15 connected clients regardles quality of connection
even in noncongested area, not one but several ap with more than 15 clients falls in that rule
its simple, overall bandwith is no more than 25Mbps, and that in a situation where customers do not make or attempt to make traffic
to be exact, when all clients a silence, and you test aggregated bandwith with 4-5 best ccq clients we can exceed magical number 25Mbps
we had that troughput with A mode and nstream in past but that was expected then
it looks like ap in a large number of mixed mode clients falling into a some kind of "safest mode" to achieve a stable connection to every client and so drop self-performance to lowest 25Mbps.
its just not god enough, we need at least 40+Mbps

vlatko · Sat Jul 23, 2016 12:14 pm

to follow up on previous post, next very irritating situation is from testing in previous noncongested ap examples is,
(noncongested aps) best 4-5 mentioned ccq clients can achive 40+Mbps agreggated troughput on ap when switch wireless on 802.11 mode, to be honest 802.11 mode degrade possible troughput for low ccq clients on this aps.
(congested aps) on congested urban aps we dont have such gains switching to 802.11 mode becouse interference and simillar problems. in these circumstances 802.11 ap is very unstable and unreliable, requesting constant monitoring. aps in nv2 mode is far more stable in this conditions.
so in this urban congested conditions we have stable performance with nv2 but with 20+Mbps troughput restrictions, or variable troughput totaly unreliable from 5 to 50Mbps with 802.11 mode.
what is our conclusion from this
we need further tweaking of nv2 from mikrotik
we are aware that mikrotik cant do much about aps in congested areas, but we would like tweaking nv2 to drew as much from aps in noncongested areas close by 802.11 troughput with nv2 stability. I think it would be possible to do so at the expense of low ccq clients simillar to 802.11 mode.
all point is next. dont evens all clients in nv2. allow best clients to derive maximum troughput regarding ccq, we deal with others (mean low ccq).

digitexwireless · Thu Nov 03, 2016 7:01 pm

Anybody tried the new wireless packages? I worked with support earlier this year and it helped setting NV2 TDMA period to Auto. Downside is sometimes during radius authentication you see connecting over and over in the log and no or only a few clients connect.

As for other manufacturers, ePMP works great and scales up over 30+ client beautifully. PMP 450 also scales up nicely.

I would love to solve this using the 1000's of Mikrotik clients I have. A forklift upgrade would not be good for the balance sheet.

mennowz · Thu Nov 03, 2016 10:31 pm

Hi everyone,

i have something different going on with our NV2 sectors. One of them as 25 clients connected, it runs 5.26 signals ranging from -36 till -60.

This sector is easily processing 40+ mbps, not joking here.. It will go to around 80/90 mbps.

What helped us immensly was fixing MCS rates, we are only using MCS 0 as basic rate and MCS12 as supported rate. Also changed period size to 4ms.

Oh yeah, radio is a RB912UAG.. just over 2 years old.

Anyone wants to test with these settings?

Menno

mennowz · Thu Nov 03, 2016 10:34 pm

Oops, forgot to mention : running on 40Mhz channels and using RF Elements CC 90 degrees sectors...

Menno

marekm · Fri Nov 04, 2016 12:59 am

Anybody tried the new wireless packages?

I did (6.37.1, AC PtMP setup in a clean licensed 6GHz channel), but don't see much difference. Like many people earlier, I can see packet loss and that's what slows down TCP throughput. UDP, or many parallel TCP sessions, can go much faster. That's just the way TCP works, even very small loss -> low speed. But we expect much higher speeds today than just a few years ago. So the loss could be considered acceptable a few years ago when NV2 was designed, but today it is not. According to the formula from https://en.wikipedia.org/wiki/TCP_tuning#Packet_loss (loss under square root) for 10 times more speed, you need 100 times lower loss. So, very low packet loss is critical today, much more than before.

As the loss is not caused by too much traffic (latency doesn't go up significantly), I suspect one (or both) reasons for this loss in NV2:
1. Too small queues (number of them is configurable in NV2 QoS settings, but size is unknown and not configurable), not big enough for high speed bursty traffic; should be much larger at today's 802.11ac speeds than it was back in the old days of 802.11a, and ideally the size should be configurable separately for each of the 2/4/8 QoS queues (optimal size varies for different kinds of traffic).
2. Too few retries (like hw-retries in 802.11, where it is 7 by default but can be set up to 15; in NV2 the number is unknown and not configurable) before giving up in presence of packet errors, which are unavoidable (even if we have a clean channel, the antenna also picks up strong signals from other channels which can overdrive the receiver and distort our signal - this is a common situation when many radios are on the same mast).

MikroTik, please - do something about it.

pukstup · Sun Mar 25, 2018 12:40 am

Hi everyone,

i have something different going on with our NV2 sectors. One of them as 25 clients connected, it runs 5.26 signals ranging from -36 till -60.

This sector is easily processing 40+ mbps, not joking here.. It will go to around 80/90 mbps.

What helped us immensly was fixing MCS rates, we are only using MCS 0 as basic rate and MCS12 as supported rate. Also changed period size to 4ms.

Oh yeah, radio is a RB912UAG.. just over 2 years old.

Anyone wants to test with these settings?

Menno

Hi Menno,

I have the same setup as you are describing here. Only difference is I'm running 2.4Ghz-only-n in the countryside, with 19 clients. There is very little frequency interference, but am not getting anywhere near 40mbps, more like 2mbps. I would like to try fixing the MCS rates, but in the HT MCS tab, the options are greyed out. I do have a level 4 license on my router. What am I missing here?

JB172 · Sun Mar 25, 2018 10:08 am

Hi everyone,

i have something different going on with our NV2 sectors. One of them as 25 clients connected, it runs 5.26 signals ranging from -36 till -60.

This sector is easily processing 40+ mbps, not joking here.. It will go to around 80/90 mbps.

What helped us immensly was fixing MCS rates, we are only using MCS 0 as basic rate and MCS12 as supported rate. Also changed period size to 4ms.

Oh yeah, radio is a RB912UAG.. just over 2 years old.

Anyone wants to test with these settings?

Menno
Hi Menno,

I have the same setup as you are describing here. Only difference is I'm running 2.4Ghz-only-n in the countryside, with 19 clients. There is very little frequency interference, but am not getting anywhere near 40mbps, more like 2mbps. I would like to try fixing the MCS rates, but in the HT MCS tab, the options are greyed out. I do have a level 4 license on my router. What am I missing here?

Hi pukstup,

Go to "Data Rates" and click configure. Unckeck all the Rates. Then go to "HT MCS"

Mon Mar 26, 2018 8:55 am

Please try to upgrade the AP to the latest v6.42RC version as we have made improvements to the Nv2 PTMP setups.
Here is the link to the topic:
viewtopic.php?f=21&t=132181

networkfudge · Mon Mar 26, 2018 9:29 am

Here's what I would do

A. Go to base level
1. Upgrade ROS to anything newer than 6.42rc46 (significant improvement in nv2)
2. Set a 20Mhz channel
3. Set all data rates to auto and remove all other non default settings / "tweaks"
4 Change nv2 period size to auto with dynamic downlink

B. Find a clean channel
Identify a few candidate channels by using the scanner and freq. usage tools. I like to keep a spare N device on my towers to use for spectral analysis but if this is not an option then just use scanner and freq. usage
Once you have a few channels that look good then start running bandwidth test to a station with good signal levels and start cycling through your candidate frequencies until you find the one with the best compromise between throughput and consistency.

C. Identify bad stations
Then use the access list to kick off all weak stations one by one starting from the worst until you find the ones that have a dramatic decrease in performance. Then find a better solution for those customers (better CPE, better position, better alignment, etc)

D. The rest
Lastly you can try tweaking other settings in wireless menu but always test to see if they have a positive impact, if not then revert back to base level.

Good luck!

pukstup · Mon Mar 26, 2018 11:26 pm

Thank you, everyone.

Your tips helped make a substantial improvement on my network.

Who is online