Looking for a GOOD example of Nstreme in multipoint

Adam McLaughlin · Mon Jan 25, 2010 7:44 pm

Hello RouterOS fans,

I am looking for a good example of a multipoint network where Nstreme w/ polling has been used successfully. We have several access points that have very large view sheds where clients may be ten miles away from each other, thus causing the classic "hidden node" problem. We are looking at polling Nstreme to alleviate this problem, yet in some of our tests we don't get the results that we expect to see.

How has this worked for you? Can other people post their succesful configurations? I am looking for Polling Nstreme access points with client counts at twenty five or higher.

Thank you,

Adam

Muqatil · Tue Jan 26, 2010 11:49 am

Hello Adam,
I'm a proud user of nstreme PTMP.
What version of ROS you used for your tests?
I would suggest using 3.30 with wireless test or 4.5 (remember to do a wireless interface reset configuration after the upgrade to 4.5).
I've the limit set to 40 clients on each wlan and each user is fed correctly with a low latency/jitter. Previously our bottleneck was the CPU of the Routerboard. Now we deploy RB411AH+R5H and they work flawlessly.

Adam McLaughlin · Tue Jan 26, 2010 6:21 pm

Hi Medianet,

Wonderful! Can you elaborate on what edition of RouterOS your clients are using? Also, what are your particular Nstreme settings?

This is with polling enabled at the Access Point, right?

Adam

16mc · Tue Jan 26, 2010 11:39 pm

We are using Nstreme PMP on all our APs serving over 600 clients. Most APs have around 30 or so clients and clients have packages ranging from 1.3 to 5mbps down load. Depending on the AP we see around 10mbps of through put per AP(10mhz channel). Most units are 3.30 with wireless test package, and a hand full of 4.x units. Most APs are 411A or AH's running an XR2 card, with a hand full of 5ghz and 900mhz units in the mix all running Ubiquiti cards. Clients are mostly R52H card for 2.4ghz ans some R5Hs for 5ghz.

We had one AP that at one point was serving 40 clients + 2 "repeaters" for over 75 clients. It has now been sectorized and all units are under 40 clients at this time.

Adam McLaughlin · Wed Jan 27, 2010 2:52 am

Hi 16MC,

That is ideally what I am trying to do; I need to implement the polling set up to alleviate the hidden node problems that we are seeing AND the streaming video factor of our Netflix addicts. Need to do something to balance the load between all of the users and make sure that no one gets shut out.

A bad link is still a bad link regardless, but we are looking to polling to level the playing field.

What settings did you find worked best for your 40 client sectors?

Adam

Wed Jan 27, 2010 10:53 am

Seems Nstreme needs all participants to have a fast cpu and the benefit
seems not as big as expected. As I've never seen any statement from MT
regarding a running example I'm wondering MT even has a outdoor test
setup for there developer(s). And without I dont believe they can implement it
efficient.

We've decided to implement RTS/CTS instead as:
- we've still a lot of RB133c out there
- it's interoperabel with other equipment and 802.11n
- you've to implement nstreme it on the whole segment in one sweep as MT has no Nstreme autodetection

Wed Jan 27, 2010 10:57 am

ste, did you read above success stories?

yes, MikroTik does have outdor Nstreme setups. Even our internet uplink for the office is running over Nstreme wireless link, so if there would be issues, everyone here would notice

Wed Jan 27, 2010 11:39 am

ste, did you read above success stories?

yes, MikroTik does have outdor Nstreme setups. Even our internet uplink for the office is running over Nstreme wireless link, so if there would be issues, everyone here would notice

Hi Normis,

I use Nstreme for ptp and small scale ptmp (up to 3)
with good signal conditions. In this setup Nstreme is very
good. I've not seen a success-story with bigger scale
(up to 40) clients with some bad signaled clients in between.
If you can provide me a link ...

Stefan

Wed Jan 27, 2010 11:59 am

the two above setups with 40 clients are not good enough examples

?

Wed Jan 27, 2010 12:48 pm

the two above setups with 40 clients are not good enough examples ?

It does not tell me:

- what happens when I've 133c3s in between
- what happens when I've clients with a -88 signal in between
- what happens when 3 unlimited users are torrenting

So it does not tell me if nstreme ptmp would work for me.

Stefan

Wed Jan 27, 2010 12:50 pm

you ask too much

but maybe 16mc and medianet can tell you?

Muqatil · Wed Jan 27, 2010 2:23 pm

Our clients have mostly RB411s, there's some 133c out there still, but we are replacing them since they arent working good with crowded radios. They work fine until the user starts downloading from emule/torrent. The 133c wont handle both Natting and polling with small packets.. The ROS Version vary from 3.2 to 3.30. Only few of them have wireless-test package enabled.

/interface wireless nstreme> pr
 0 name="wlan1" enable-nstreme=yes enable-polling=yes disable-csma=no 
   framer-policy=dynamic-size framer-limit=2000

This is my nstreme configuration. The framer-limit is quite useless since i noticed the frame is always 1532 (pppoe encapsulation).. With exact-size you might be able to get more throughput but the latency might be higher.

It does not tell me:

- what happens when I've clients with a -88 signal in between
- what happens when 3 unlimited users are torrenting

So it does not tell me if nstreme ptmp would work for me.

Stefan

With wireless-test, nstreme isolates the bad client with high signals.. (our clients disconnect @ -84). That protects the other "good" clients.
We havent unlimited users, and we shape peer-to-peer traffic to keep the quality of the links fair enough.

This of course turns out to be bound to mikrotik, and we cannot buy for example any ubiquiti clients. But until now we are fine with Mikrotik, even if sometimes the new releases are fucked up

16mc · Wed Jan 27, 2010 5:04 pm

In most cases we use the default packet size of 3200 with best fit, with CSMA disabled(box checked). Hardware retries are set to 6. Adaptive Noise Immunity is turned on. We find we have to do some fine tuning on some sites to do to local conditions. If the AP has issues (many client disconnects) we will increase the Hrdwr retries to 8 and drop the the max packet size to 1500, this helps in preventing corruption of larger packets.. you also loose a bit of through put. CSMA in early 3.x seem to make things worse particularly if the RF was not great. Later v3.x and 4.x it seems to help. We also fine tune the customers end by adjusting there modulation rate to a level that provides a good CCQ and reduces/eliminates constant changing of data rates. This provides a better connection and helps reduce latency and jitter. In most cases we use 6(basic) 24,36mbps and add in 18 or 48 as dictated by the link. You can also get pretty good results by selecting Modulations at the AP too. What works best will depend on your installation. We also make an effort to keep all clients above a -70 signal strength, and don't hesitate to correct problem clients.

We have not directly compared our set up to using RTS/CTS but a polling system should out perform it. While RTS/CTS helps a lot compared to a system that does not use it. It is still a "collision" based and would work best with smaller more compact (area wise) configuration.

One thing we would like to see and is on the top of the MT wish list (wiki) is gps syncing. While it is not a fix all it seems to be on the surface it would help a lot at densely packed sites and keep cost down as few filters would be needed, in wide open areas where were an omni can be seen from another tower more then 50km away.........HINT...HINT...NORMIS!

netrat · Wed Jan 27, 2010 5:38 pm

Our clients have mostly RB411s, there's some 133c out there still, but we are replacing them since they arent working good with crowded radios. They work fine until the user starts downloading from emule/torrent. The 133c wont handle both Natting and polling with small packets.. The ROS Version vary from 3.2 to 3.30. Only few of them have wireless-test package enabled.

Why is your 133c doing NAT?

Your AP or PPPoE concentrator should be taking care of NAT.

Wed Jan 27, 2010 5:56 pm

Our clients have mostly RB411s, there's some 133c out there still, but we are replacing them since they arent working good with crowded radios. They work fine until the user starts downloading from emule/torrent. The 133c wont handle both Natting and polling with small packets.. The ROS Version vary from 3.2 to 3.30. Only few of them have wireless-test package enabled.
Code: Select all
/interface wireless nstreme> pr
 0 name="wlan1" enable-nstreme=yes enable-polling=yes disable-csma=no 
   framer-policy=dynamic-size framer-limit=2000 
This is my nstreme configuration. The framer-limit is quite useless since i noticed the frame is always 1532 (pppoe encapsulation).. With exact-size you might be able to get more throughput but the latency might be higher.
It does not tell me:

- what happens when I've clients with a -88 signal in between
- what happens when 3 unlimited users are torrenting

So it does not tell me if nstreme ptmp would work for me.

Stefan
With wireless-test, nstreme isolates the bad client with high signals.. (our clients disconnect @ -84). That protects the other "good" clients.
We havent unlimited users, and we shape peer-to-peer traffic to keep the quality of the links fair enough.

This of course turns out to be bound to mikrotik, and we cannot buy for example any ubiquiti clients. But until now we are fine with Mikrotik, even if sometimes the new releases are fucked up

We've difficulties to bring all of our clients to good signals. As we respect ETSI regulations this is difficult to do in all cases. There are a lot of customers accepting low bandwidth in Regions where they get max 64kbps otherwise.
So we've to handle it.
At the moment we're on the way to 3.30 wireless-test
and rts. We use only MT as clients but a lot of 133c.
We dont want to replace 133cs at the moment as may be soon 11n
is a gamechanger and then we need 411ahs with other antennas.

Wed Jan 27, 2010 6:10 pm

We have not directly compared our set up to using RTS/CTS but a polling system should out perform it. While RTS/CTS helps a lot compared to a system that does not use it. It is still a "collision" based and would work best with smaller more compact (area wise) configuration.

This depends on the implementation of polling, traffic pattern and signal strengths.
If you've to poll 30 weak clients while only 5 with good signal want to send...
I've not seen the polling algorithmus so I dont know if it meets my situation.

If MT implements a nstreme autodetect for clients (they dont as I've already asked) I would give it a try as I only have to change it at the AP.

Adam McLaughlin · Wed Jan 27, 2010 6:51 pm

This is some wonderful feedback and a great discussion; just what I was looking for.

We limit our client signals to -75; if they can't make this threshold then we don't install them. We find that this levels the playing field dramatically.

We don't see bittorrent as much as we have problems with users running constant movie downloads; anything that keeps the A.P. busy for a very long time is bad for business. We're looking to polling with some tuning to help alleviate the case where people may be locked out or complain or slowness with high latency.

Adam

Adam McLaughlin · Thu Jan 28, 2010 9:53 pm

One thing we would like to see and is on the top of the MT wish list (wiki) is gps syncing. While it is not a fix all it seems to be on the surface it would help a lot at densely packed sites and keep cost down as few filters would be needed, in wide open areas where were an omni can be seen from another tower more then 50km away.........HINT...HINT...NORMIS!

I agree, I too would like to see GPS derived client management. If there is any way that I can assist with this, I would like to volunteer.

Adam

RAket · Tue Feb 09, 2010 3:00 pm

Seems Nstreme needs all participants to have a fast cpu and the benefit
seems not as big as expected. As I've never seen any statement from MT
regarding a running example I'm wondering MT even has a outdoor test
setup for there developer(s). And without I dont believe they can implement it
efficient.

We've decided to implement RTS/CTS instead as:
- we've still a lot of RB133c out there
- it's interoperabel with other equipment and 802.11n
- you've to implement nstreme it on the whole segment in one sweep as MT has no Nstreme autodetection

I think you've made good decision. We test and use wireless-test package on one site with 10-15 clients per sector. But: 133c rb's are slow for polling and nstreme. Clients must have good signals, if they don't, cpe's are disconnecting. You have to use RB433AH or better hw on AP. Latency is not so good for me. It is from 5-50ms.
Positives are that you can have good throughput on sector (24-28Mb/s, TCP). It solves hidden node problem. Eg. five clients downloading 5Mb/s in the same time.

I want to test rts/cts on some AP where we have old hw and non-mikrotik cpe's. Maybe I will post some results in the future.

Adam McLaughlin · Tue Feb 09, 2010 6:32 pm

Please post some results with your test setup; I would be very enthusiastic to hear what you find.

Adam

znet · Tue Feb 09, 2010 7:12 pm

In most cases we use the default packet size of 3200 with best fit, with CSMA disabled(box checked). Hardware retries are set to 6. Adaptive Noise Immunity is turned on. We find we have to do some fine tuning on some sites to do to local conditions. If the AP has issues (many client disconnects) we will increase the Hrdwr retries to 8 and drop the the max packet size to 1500, this helps in preventing corruption of larger packets.. you also loose a bit of through put. CSMA in early 3.x seem to make things worse particularly if the RF was not great.

Why would you ever disable CSMA in PtMP environment? A backhaul with a clear channel and good SNR might benefit from disabling CSMA, but in any other condition it would seem to be counterproductive, and exacerbate problems, especially with APs that have clients with significantly different signal strengths.

BTW, I havent had any problems with NStreme, and in general have never seen good performance without it.

Carrier Sense Multiple Access (CSMA) should be self explanatory. No CSMA, the clients have a free-for-all and transmit whenever they want. My experience is leave it on. Have only seen marginal improvement on strong backhaul links (PTP).

Ah, yes. If GPS sync was implemented across the board as suggested (like many other types of carrier grade eqpt), it would go a long way for stability. This would be especially true for cell-sites that have many radios transmitting in the same band, e.g. 12 APs and 6 backhauls in the same band on the same roof--takes a big roof of course...

Thats my story, and Im stickin to it.

Adam McLaughlin · Tue Feb 09, 2010 7:30 pm

Ah, yes. If GPS sync was implemented across the board as suggested (like many other types of carrier grade eqpt), it would go a long way for stability. This would be especially true for cell-sites that have many radios transmitting in the same band, e.g. 12 APs and 6 backhauls in the same band on the same roof--takes a big roof of course...

Thats my story, and Im stickin to it.

We too would love to see some more development with GPS sync for stability!! ( Hint Hint )

Adam

znet · Wed Feb 10, 2010 8:26 am

Adam was looking for a good example of NStreme in a PTMP environment. I still dont understand why NStreme is so misunderstood. I can only speak from years of doing it in a dense urban environment where when scanning I can detect dozens and dozens of APs in the band, with narrow antennas, its NStreme that takes the wiley client and tells it who is boss, and when its their turn to speak. It creates as much order out of the chaos as possible. Granted, my terrain and air superiority by height helps a bit. I can assure you I have dozens and dozens of APs that are happily working for years, with dozens and dozens of clients behaving well, and playing well with the other CPEs and Boss AP

I would look to other factors regarding the client side when you think Nstreme isnt working for you. Possibly it has the effect of separating the good from the bad. If links become problematic with Nstreme, fix the link, dont disable Nstreme. The AP knows better, and so it should be. Otherwise you are chasing the wrong problem....IMHO....

Adam McLaughlin · Wed Feb 10, 2010 6:46 pm

Thank you Znet for the information.

I have one problematic site that did not take well to the polling mode, I think that this is causal to the site being very close to several cellular sites in a very rural area. From the Access Point location I can see with my eye at least four commercial cell sites operating on 881, 780 and 1.95 GHz. I did get some 8 pole band pass filters ordered, and am optimistic that this will help alleviate the odd behavior that we are seeing in this neighborhood.

Some of the XR2 clients in this area report noise floor values of -85 WAY out in the middle of no-where when their radios "scan" and see only our own Access Points, adding the band pass filter drops this value to -102.

We are looking to a Polling / Nstreme setup to allow the Access Point to manage out clients. Our towers are also the "classic" tower design, and see a lot of geographical area which means that almost all of our clients are hidden nodes to each other. This is why we are pushing ahead with a polling setup to equalize the playing field.

What I would like to know is the mechanics of how one marginal link client would affect the polling distribution; i.e. if 19 / 20 clients sigs are all -62. and one guy has a -80 signal, how does this affect the aggregate? Does the poller trip on this customer when the Access Point runs though the client registration table?

Adam

16mc · Thu Feb 11, 2010 9:42 pm

I have tried with and with out CSMA and in most cases it seems to work better with out. My take is that CSMA, being a collision detection protocol should not be needed in in a polling environment and thus should disabled as you should not have any collisions to worry about. If left on I would guess (I too would like to know more about Nstreme antilogarithm too) that it could detect regular 802.11 a/b/g and would back off, ie play nicer with other gear. Anyone with more knowledge on subject please chime in.

netrat · Thu Feb 11, 2010 10:09 pm

I was under the same impression. If you are using polling why the need for CSMA?

Adam McLaughlin · Thu Feb 11, 2010 11:07 pm

Here is what we are dealing with:

Five years ago, someone would come home from work and check their email, then go and sit in front of the TV for the rest of the night.

Now, people come home from work, they turn on the computer and expect to watch streaming video, be it movies or television shows. They want to use P2P software to transfer material that they don't want to pay for, or can't get through conventional approaches. They might use their email to send someone a note, but the popular thing for today's subscriber is some kind of video chat.

Internet usage will only go up from this point forward. It will not go down. There will be a point in the future when EVERYTHING that can be transferred via the web will be done this way. Every TV show, movie, music, etc. will be delivered to people via the internet.

We need to develop transfer methods, protocols and access point strategies to address this now so that we aren't caught un-prepared. We need to find the right configuration to serve all clients equally, regardless of their type of traffic. We have converted all of our non-RouterOS clients to RouterOS clients to enable their compatibility with Polling Nstreme, now we need to find the best recipe for success using this to serve the clients as equally as possible.

I have heard nothing but good things about the GPS synchronization of the access point / client Time Domain Division strategy that other wireless networks have utilized, and would be thrilled to see this in RouterOS.

( I am volunteering to BETA test this Normis!! )

Adam

16mc · Thu Feb 11, 2010 11:30 pm

GPS syncing is good but it is not all roses. While I have not personally used some of the equipment that does I have heard on several occasions that that you do run in to some issues, particularly if you cells are not close in size. Guessing it has to do with the time to reach the farthest client on a big cell is not a good timing sequence for a small cell.

Either way it would make life a lot simpler as we have some sites that are "full" in terms of spectrum and we could squeeze some more in with GPS SYNC. A good channel filters can work wonders against most issues and should not be over looked. Not only can you provide separation from you own gear you protect your self from other gear too. Some say with filters you can't change channels in case of interference.....I say I've have yet to need to change a channel that is filtered. Many products like Moto have to have the sync as the RF and antennas are so bad you could not collocate with out syncing. Nothing like a solid RF system to start with

MT bring on the sync!

Adam McLaughlin · Thu Feb 11, 2010 11:40 pm

We don't install or set up a site with out at the VERY LEAST an 8 pole band pass filter on the antenna leads, sometimes a band pass filter with a notch depending on what we are working with. We are very conscious about co-location with other services and the facts of life of doing so in an RF hot environment.

My college job through Electrical Engineering program at Cal was for a communication site development, management and troubleshooting company, I have 10+ years experience at that hair pulling job.

What we need is a very reliable method to guarantee uniform access to clients regardless of traffic use. The RF management component in terms of freq selection and interference is up to the guy managing the ISP, what we want is for RouterOS to do their part.

Adam

16mc · Thu Feb 11, 2010 11:52 pm

Couldn't agree with you more. We like the platform and its flexibility as it allows us to do things the competition can't. And we can do it all on the same interface, which helps in other areas. Worked a lot with Waverider over the years and was fairly impressed with there polling. Great distributions over the clients, and when you started to run out of bandwidth it would start slowing the clients (4 levels) from the bottom up so your low end customers would get slowed first... Part of this was to poll the low level clients less and the higher ones more, worked very well and never had many problems with hogging of the bandwidth. Would be nice to keep seeing big improvements in Nstreme. A road map would be nice as it would help me plan for the future.... right now I still have to keep hoping the development goes were we need it to!

Looking for a GOOD example of Nstreme in multipoint

Looking for a GOOD example of Nstreme in multipoint

Re: Looking for a GOOD example of Nstreme in multipoint

Re: Looking for a GOOD example of Nstreme in multipoint

Re: Looking for a GOOD example of Nstreme in multipoint

Re: Looking for a GOOD example of Nstreme in multipoint

Re: Looking for a GOOD example of Nstreme in multipoint

Re: Looking for a GOOD example of Nstreme in multipoint

Re: Looking for a GOOD example of Nstreme in multipoint

Re: Looking for a GOOD example of Nstreme in multipoint

Re: Looking for a GOOD example of Nstreme in multipoint

Re: Looking for a GOOD example of Nstreme in multipoint

Re: Looking for a GOOD example of Nstreme in multipoint

Re: Looking for a GOOD example of Nstreme in multipoint

Re: Looking for a GOOD example of Nstreme in multipoint

Re: Looking for a GOOD example of Nstreme in multipoint

Re: Looking for a GOOD example of Nstreme in multipoint

Re: Looking for a GOOD example of Nstreme in multipoint

Re: Looking for a GOOD example of Nstreme in multipoint

Re: Looking for a GOOD example of Nstreme in multipoint

Re: Looking for a GOOD example of Nstreme in multipoint

Re: Looking for a GOOD example of Nstreme in multipoint

Re: Looking for a GOOD example of Nstreme in multipoint

Re: Looking for a GOOD example of Nstreme in multipoint

Re: Looking for a GOOD example of Nstreme in multipoint

Re: Looking for a GOOD example of Nstreme in multipoint

Re: Looking for a GOOD example of Nstreme in multipoint

Re: Looking for a GOOD example of Nstreme in multipoint

Re: Looking for a GOOD example of Nstreme in multipoint

Re: Looking for a GOOD example of Nstreme in multipoint

Re: Looking for a GOOD example of Nstreme in multipoint

Who is online