PDA

View Full Version : Mesh network


junkie
09-15-2009, 11:46 AM
Hey guys, looking to set up a mesh network ni a campgroung, I am not too farmiliar with mesh, so looking at the best place to start as far as the hardware goes. I read on the Mikrotik forums that Mesh doesn't really work too well with Router OS, was thingking about 3 cards per node, 1 to send 1 to receive and 1 as the AP fo rthe end user...

any help would be appreciated.

regards, Mark

WHT
09-15-2009, 12:02 PM
I wouldn't go with any mesh. I'd go with a centrally mounted, or highest placed location with an omni antenna using a Powerstation 2-EXT and scatter some Pico 2 or Bullet 2 (with less expensive indoor 6 dBi antenna and stick it all in a capped PVC pipe), then run those radios in WDS mode all peered back to the central AP.

junkie
09-15-2009, 12:56 PM
Thanks, I was thingking mesh because not all the 'nodes' will be able to see the central AP, so 1 node might have to go through another node to get to the central AP. if set up in WDS mode, wouldn't the bandwidth take a big hit going through multiple WDS repeaters to get back to the central AP? I would also like to keep the backhaul in the 5GHz band, and the AP's for the end user in 2.4.

Mark

rconaway
09-15-2009, 06:09 PM
The ideal situation is that you enter a SSID in a radio, it meshes up with whatever radios that are on the same SSID, and life is perfect. In reality, mesh doesn't really work that well and has very little advantage over a properly designed PTMP to PTMP design. Use WDS for your backhauls, keep your radio in AP+WDS and you are good.

junkie
09-16-2009, 11:14 AM
what kind of bandwidth hit would I take for using WDS and going through say 2 hops to get back to the central AP? and I have never dealt with radios this close together, so channel selection is up in the air, are all the backhaul radios one 1 ch. and all the WDS AP's on another Ch.? not seperate ch.'s for each radio correct?

rconaway
09-16-2009, 12:16 PM
what kind of bandwidth hit would I take for using WDS and going through say 2 hops to get back to the central AP? and I have never dealt with radios this close together, so channel selection is up in the air, are all the backhaul radios one 1 ch. and all the WDS AP's on another Ch.? not seperate ch.'s for each radio correct?

All radios have to be on the same channel. Your second hop takes a 50% performance hit if the modulation rates are the same between hops. You want to try and modify the power output to minimize the third hop. Consider 2 radios going back to 1 instead of a hop system if you can.

junkie
09-16-2009, 01:57 PM
For sure, I am going to try to limit all the AP's to 1 hop if at all possible, just wondering what the hit would be for say 2 hops back to the AP... I would imagine that all the AP's would need the same SSID as well correct; so that the end users could roam about and not have to constantly choose the AP with the strongest signal?

thanks again.

WHT
09-16-2009, 02:09 PM
so that the end users could roam about and not have to constantly choose the AP with the strongest signal?That changes things...like adding 500% to your project cost for fast handoff features.

junkie
09-16-2009, 02:22 PM
ah ok then... not that having them reconnect is that bog of a deal; 99% of the time the end user will be stationary, and f they do get disconencted because they changed locations, connecting to another AP isn't that huge...

thanks.

rconaway
09-16-2009, 02:31 PM
One hop isn't a big deal. Fast handoff is nice and I think it's being looked at. I'm a big fan of it.

junkie
09-16-2009, 02:35 PM
The ideal situation is that you enter a SSID in a radio, it meshes up with whatever radios that are on the same SSID, and life is perfect. In reality, mesh doesn't really work that well.

Can you elaborate on this, is it a fatal flaw in the mesh protocol, hardware, or a little of both?

thanks,

rconaway
09-16-2009, 04:17 PM
This could be a very long discussion but I'll start. Every commercial version of mesh implementation is different and is also based on the unique flavor of hardware they supported. The second issue involves the concept of mesh. Is it just finding the closest radio? Is it based on load-balancing? Does it use a single 5.8GHz radio for backhaul and 2.4GHz for area coverage? Does it use multiple radios for backhaul, etc.... There are a million questions.

Taking it another step, the complexity of the networks varies. For some stupid reason, since most firmware is Linux based, the management servers have to be Linux instead of Windows making it harder to find and train people for support. Some vendors dont' believe in layer 2, they stay in layer 3. Unless they need layer 2 so they put some fake layer 2 structure in there making it excessively complicated and much harder to troubleshoot. Oh wait, let's throw in features that you really can't use in the real world but let's not tell the VARs those features shouldn't be used because the CPU's aren't powerful enough.

Throw in firmware upgrades that basically disconnect all the radios so that you have to get a lift to access evey AP in the city to install new firmware. That's a budget buster. Market UDP traffic numbers, not real-world TCP/IP numbers along with under guessing how many AP's per square mile. Now, assume you are spending $5000 per year or more for support, along with AP costs from $1000-$11000 per AP, etc.....

Okay, you wanted a more technical answer. After learning all those things the last few years, I have decided I want simple. Yes, there are places that these systems work great at $100-$200K per square mile. Some are great at long range, others at fast handoff, yet others handle video very well. If I'm supporting it though, I want simple.

WHT
09-16-2009, 06:05 PM
"In reality, mesh doesn't really work that well."

Look at all the MuNi WiFi city wide mesh projects that have failed.

Why?
* Its not scalable.
* It costs far too much to expect pushed advertisements as the sole profit model to pay for itself.
* Those pretty pictures of a mesh node connect to all those other mesh nodes is a myth. In reality each node would only be able to see one or two, perhaps three of its nearest nodes, and even then they have to be within a thousand feet or so. Remember, you're working with typically a 6 dBi antenna or a max EIRP or 36 dBm at high data speeds.
* Poor basic RF planning created 200% cost overruns.

However....With the low cost per bandwidth of the new Ubiquiti line, that could bring the cost down to under $3,000 per square mile. Compare that to mesh networks costing $50,000 per square mile (for comparable performance).

What *DOES* work is a scattering of low cost 2.4 radios that all backhaull via 5.8 to centralized locations, perhaps one sub AP controller per a section of town town.

junkie
09-16-2009, 08:05 PM
Thanks for the advice... it's greatly appreciated.

rconaway
09-16-2009, 10:08 PM
Justin was far more elegant and organized than I was but you get the gist. In most mesh networks, each radio can only see forward and back. In that case, what good is mesh. Now Fast Handoff, that's something I would support :).

WHT
09-16-2009, 11:24 PM
Sorry, I may have inadvertently hijacked the thread by bringing up Muni WiFi where the big time players estimate:
* Chicago - 220 square miles, 7,500 access points, $18.5 Million or 34 APs per sq. mile
* Google's Mountain View - 12 square miles, 400 access points or 33 APs per sq. mile
* Earthlink Houston - 640 square miles, $50 Million. Closed shop and city took it over for meter reading.
* Earthlink anywhere
* Corpus Christi - 147 square miles, $7.1 Million

In spite of failed engineering (proposing solar power in areas of tall buildings that would be in the shade most the time, all the hotspots and SOHO wireless routers that would bung up the works, and the list goes on...), the payback model was based on click-through advertising revenues (think about, how often do YOU buy something based on an ad you see on CNN, Foxnews, MSN, or Yahoo?..and advertising nag factor that they assumed would compel you to just pay for the dammmed service).

Nevertheless, mesh has indeed been successful in several very small one by one mile small town deployments for coverage in core "downtown" areas for low-bandwidth users that need absolute mobility? Perhaps when laptops are given out at the local homeless shelter, you may see the need for mobility. Marketing idea! Front seat car mounts for laptops that fit a shopping cart!!!

I mean...how much can you download sitting on a park bench. And just how many business class users are going to need that connection when they can go back inside their glass corner office and get a "real" connection. And think about it...when was the last time you even *saw* a park bench...and if you did, it was prolly war-driving marked with wet paint.

And *THAT* is my "elegant and organized" opinion, though I prefer "well jaded".

WHT
09-16-2009, 11:28 PM
* Chicago - $84,000 per square mile
* Earthlink Houston - $78,000 per square mile
* Corpus Christi - $48,000 per square mile

rconaway
09-17-2009, 08:28 AM
Actually, Justin's numbers don't include the amount of money it took to install the backhaul and management infrastructure not to mention the on-going expense. Corpus Christi has 1 full-time and 10 part-time IT people, not to mention the cost of physically supporting and repairing the network. In exchange however, they have the majority of their water meters, 55,000 automatically read. That would normally save about a million dollars a year. If they had drive by systems, that would save about $250K per yer. However, they also have parking meters and other substructures on the network.

sxpert
09-18-2009, 07:05 AM
For some stupid reason, since most firmware is Linux based, the management servers have to be Linux instead of Windows making it harder to find and train people for support.

That utter bull**, FUD and Troll all mucked in one sentence, and you know it !

Steve ballmer, come out of that corpse !

Besides, I don't see why, while the firmware in the wifi device would be linux based, there couldn't be a windows based management software, or even better, some cross platform implementation of such

WHT
09-18-2009, 07:46 AM
I am so watching this thread.......

rconaway
09-18-2009, 08:37 AM
Okay, Linux is the greatest thing on earth. Command line operating systems rock. Now, grab any 10 computer people and tell me how many of them could install Windows and how many of them could install Linux? Then ask them something simple like, "look up the MAC address of the network cards".

Of course, Linux installs itself depending on which servers I have. After having to install Debian Linux on 2 brand new HP servers last year and can't get it installed through CD-ROM or USB keys, because of CD-ROM driver incompatibilities, and oh by the way, has no drivers from HP to solve the problem while I'm on-site 300 miles from home in front of the customer billing them hourly, I kind of decided that Linux really wasn't ready for all applications yet. That little event cost me over $2000 and made my customer very, very happy. BTW, I no longer have that customer and this little event didn't help. Can't say I ever had that problem in Windows.

It takes hours or at worst case a couple days to train someone on setting up and managing a windows servers. I have had 3 Linux consultants work for me, one of who was the president of the Phoenix Linux group and 2 more who I paid for through temp agencies, that couldn't even give me straight answers when asked for specific information and configurations. If the networks involved multiple subnets, they started babbling bs because the couldn't figure out the issues. We ended up tossing all Linux software except for one program, got our money back from the temp agencies for these so-called consultants, trained our windows guys with enough knowledge to set up and install a server for this one program, and decided we will not use Linux again internally unless there is no other option on the planet.

It might work for some people but there just aren't enough real experts out there with formal training to support it. In the real world, I want something I can install in 2 hours from scratch without re-typing War and Peace to configure it.

WHT
09-18-2009, 09:00 AM
Considering it took me only 6 hours to read War and Peace, but 12 hours to read The Agony and Ecstasy...I'd say it was more compared...oh, nevermind.

rconaway
09-18-2009, 10:38 AM
I didn't know they made a comic book version of War and Peace.

shapiros
09-18-2009, 11:29 AM
If you two were not so good at this, you'd be DANGEROUS (the general public may still be in jeopardy with "ya'll" out there, somewhere)!

WHT
09-18-2009, 01:50 PM
I didn't know they made a comic book version of War and Peace.No, they made a Windows version of it.

rconaway
09-18-2009, 10:03 PM
That would be the windows mobile version. Of course, if it works the way my windows mobile phone does, the book shuts in my face everytime it downloads a text message.

WHT
09-18-2009, 10:37 PM
Ya shuda got the Palm Treo Pro.......

macosoft
11-16-2009, 12:11 PM
Spanning Tree Protocol should be enabled on AP WDS to AP WDS?

I have 2 AP WDS and after a few hours I get a network loop. Still not figure where it comes from?!

rconaway
11-16-2009, 01:28 PM
I have never seen this problem. It has to be coming from something else.

CzechEnglishFrenchGermanItalianPolishPortugueseRussianSpanish
vBET 3.5.4 gives automatic translations