[Commotion-discuss] Building a commotion network to support 2000 concurrent users

[Ben West]

Hi Paul, likely the USB driver will remain a potential bottleneck, as your
USB dongle itself will presumably only talk at USB 2.0 speeds and not
utilize 3.0 anyway.  You would need a dongle that is advertised at USB 3.0
compliant, and which has robust linux / OpenWRT support.  My guess is that
mini-PCI would probably offer better options for adequate bus speed +
sufficient driver support.


On Sun, May 11, 2014 at 12:10 PM, Paul Liu <paulsc.liu at gmail.com> wrote:

> Hi Ben,
>
> Thanks for the suggestion. I will check them out. I do like to ask your
> opinion about the following:
>
> Several new ARM boards with 8 cores that is coming out have USB 3.0
> build-in. On-board memory will start from 2GB and some might have 4GB. Do
> you think USB 3.0 WiFi dongle with 8 core ARM board might provide good
> performance?
>
> We will get some WiFi dongle with big antenna or modify it and graft some
> big antenna onto it.
>
> Driver might be an issue. We do not know until we have the board and WiFi
> dongle to run some test.
>
> Paul
>
>
>
> On Mon, May 12, 2014 at 12:15 AM, Ben West <ben at gowasabi.net> wrote:
>
>> Wow, that would be a lot of CPU and RAM behind a single 802.11n radio.
>> Given at host machine with that many resources, I think you may find the
>> USB dongle itself becoming a bottleneck far before the host CPU sees any
>> significant load.  The USB protocol stack itself (e.g. the ath9k_htc
>> driver) is going to impose limitations on throughput in and out of the
>> dongle, likely to be far less than the USB link speed.  Embedded nodes use
>> faster bus interfaces such like mini-PCI for this reason.
>>
>> Also, you would be sacrificing some channel capacity if your USB dongle
>> has only 1 RP-SMA port (making 2x2 MIMO unavailable), or if it uses itty
>> bitty internal patch antennae.
>>
>> For a node capable of hosting 30-50 clients, here is an ARM9-based single
>> board computer, with a mini-PCI slot:
>> http://www.gateworks.com/product/item/laguna-gw2388-4-network-processor
>>
>> Likewise, here is a mini-PCI 802.11n radio from UBNT.  Two coaxial jacks
>> for antenna diversity:
>> http://shop.bizsyscon.com/ubiquiti-sr71-12-mimo-802-11b-g-n-2ghz-mini-pci/
>>
>> This bug ticket does suggest at least one person has met moderate success
>> running both of these units together under OpenWRT:
>> https://dev.openwrt.org/ticket/13465
>>
>> Also, there should be other options for single board computer  + mini-PCI
>> with OpenWRT support out there, if that is your host OS of choice.
>>
>>
>>
>> On Sat, May 10, 2014 at 2:30 AM, Paul Liu <paulsc.liu at gmail.com> wrote:
>>
>>> Hi Ben,
>>>
>>> Thank you for your feedback. We will review our overall deployment
>>> design.
>>>
>>> Here is some additional information. We will build a customized node.
>>> The prototype is using a dual core ARM cpu with 2GB ram, gigabyte Ethernet
>>> as wired backhaul. We will hookup USB WiFi dongle to service the
>>> clients.
>>>
>>> We do plan to test wireless backhaul and probably run it on 5.8GHz
>>> radios.
>>>
>>> Right now the design call for each node to support 30 to 50 clients. We
>>> will try to find out what kind of usage scenario is possible. Data only?
>>> with voice? with video? How will transfer large data affect the performance?
>>>
>>> Cheers,
>>> Paul
>>>
>>> If it is necessary, we can use quad cores or 8 cores ARM CPU.
>>>
>>>
>>> On Sat, May 10, 2014 at 1:10 AM, Ben West <ben at gowasabi.net> wrote:
>>>
>>>> My recommendation, from operating a mesh network with firmware very
>>>> similar to Commotion, that serves ~100 people daily, would be to focus on
>>>> how many mesh nodes you would need to deploy.  Likewise, since there is no
>>>> way to fit 2000 users on a single 2.4GHz channel, you would be needing to
>>>> segment such a large deployment into pocket meshes, e.g. a handful of nodes
>>>> in each mesh, with adjacent meshes programmed to non-overlapping channels.
>>>> Backhaul between these pocket meshes could be done with 5.8GHz radios, or
>>>> wired backhaul depending on your situation.
>>>>
>>>> To help with estimating how many nodes you'd need, it's fair to assume
>>>> each node could only support up to 5 (or possibly 10) simultaneous clients
>>>> before either the node itself exhausts its RAM (only 32Mbytes onboard
>>>> memory) or before the particular channel used by that ( node + its
>>>> neighboring nodes + all clients ) becomes saturated.
>>>>
>>>>
>>>>
>>>> On Fri, May 9, 2014 at 11:58 AM, Paul Liu <paulsc.liu at gmail.com> wrote:
>>>>
>>>>> Has anyone had the experience of building a commotion network that can
>>>>> support 2000 concurrent users?
>>>>>
>>>>> Any advice on how to build it, potential pit fall? total equipment
>>>>> cost?
>>>>>
>>>>> Best Regards
>>>>> Paul Liu
>>>>>
>>>>> _______________________________________________
>>>>> Commotion-discuss mailing list
>>>>> Commotion-discuss at lists.chambana.net
>>>>> https://lists.chambana.net/mailman/listinfo/commotion-discuss
>>>>>
>>>>>
>>>>
>>>>
>>>> --
>>>> Ben West
>>>> http://gowasabi.net
>>>> ben at gowasabi.net
>>>> 314-246-9434
>>>>
>>>
>>>
>>
>>
>> --
>> Ben West
>> http://gowasabi.net
>> ben at gowasabi.net
>> 314-246-9434
>>
>
>


-- 
Ben West
http://gowasabi.net
ben at gowasabi.net
314-246-9434
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.chambana.net/pipermail/commotion-discuss/attachments/20140511/ef9d8b22/attachment-0001.html>