Show Sidebar Log in

Bristol Wireless Network Report

Bristol Wireless Network Report

Introduction:
Bristol Wireless currently has 2 main operating nodes running a distribution of Debian Linux. The nodes are situated at the Easton Community Centre (ECC) and the Chelsea Inn (Chelsea). The node at the Chelsea talks back to the main node at the ECC and also runs as an open access point allowing clients to connect in and extend the network range.

Background info:
802.11b standard is 11mbit with roughly 6mbit maximum data transfer throughput.
802.11g standard is 54mbit with roughly 20mbit maximum data transfer throughput.
Most/all 802.11g access points are backwards compatible with 802.11b clients.

There are ready-made access points available which do wireless repeating via wds, but as far as I know do not allow more then one wds connection per unit.

Hardware:
Current nodes are based on DEC multias which are p100 based systems with 32mb ram. They have on board video, network and PCMCIA controllers. We are using Netgear MA311ge PCI wireless cards in these which are based on the prism2.5 chipset by Intersil and have a lot of features enabled in Linux. We are using UMR67 antenna cable runs with a selection of antennas from home made sectors (directional antenna) to purchased omni’s (omni directional antenna)

Clients connected to the network can use any 802.11b wifi compliant device to connect although we have mainly been using the buffalo wifi cards. This is due to their good Linux support and external antenna connectors. The antennas we have used for clients vary depending on their location to a node. We have used home-made waveguides (high gain directional antenna) to quarter wave omni’s (omni directional antenna).

Node Software:
As I mentioned earlier the nodes run a distribution of Debian Linux with daemons such
as zebra (routing daemon) and dhcpd (IP leasing daemon). We use the hostap wireless drivers for the prism based chipset as this enables a lot of extra features such as running in access point mode and wds mode. The Nodes basically run in ap mode and wds mode at the same time allowing clients to connect in as well as backbone links between different ap’s. We use zebra to automatically create and update the routing tables between the different types of links (ap + wds), this allows every machine on the network to be reachable by any other machine without any network address translation (NAT).

Advantages of this network structure:
The main advantage of this network structure is that we can string nodes together and increase the range of the network. As long as you can see a node which can see another node and so on you can be part of the network.

Disadvantages of this network structure:
The repeating nature of this network structure is sound in theory but leaves a lot to be desired in practice. The nature of the 802.11b wifi standard puts a crimp on the whole thing due to being half duplex, their for losing a lot of bandwidth per wireless hop. Due to the nature of RF communications full duplex systems would need 2 radio transmitter circuits running on different frequency’s to allow full duplex operation which works out very costly.

The other obvious disadvantage is that if a node goes down any other nodes beyond that node will lose connectivity and create two separate wifi coverage areas instead of one larger one.

Advantages of our node setup:
The dec mutlias are slimline desktop pc’s and don’t use that much power keeping the running costs down. They also have dual pcmcia slots and a pci slot making them very expandable. They can run a standard distribution of linux making them very customisable. They can easily be wall mounted and don’t create much noise.

Disadvantages of our node setup:
Although the multias are slimline desktop machines they still require mains input and are too large to be mounted on a roof. This means that you need a long cable coming from the antenna on the roof down to the back of the multias wireless card.

The current cable we are using is uniradio umr67 which has quite a high signal loss per meter, this limits the range of the network. The solution to this is to use high quality cable such as times microwave lmr400, but both umr67 and lmr400 are expensive to buy (lmr400 being more expensive due to better quality) This cable is also quite large, 11mm in diameter, and hard to run around the house easily.

Proposed New Network Structure:
The idea behind the new structure is to have one well positioned 802.11g (we will use 802.11b for testing) access point mounted in a box just below a high gain omni on a long pole, this will keep the cable loss to a minimum and provide maximum range.

We will lose the ability to wirelessly repeat but as long as the access point is in a good position we should have no problem with connecting and the range will be much larger. There is the idea of creating a repeating node based on the same software we currently use with a soekris board in a box below the antenna but this works out very expensive.

There are also devices called wireless media converters which basically connect wirelessly (as a client) to an access point and bridge to a wired connection. These would be perfect for clients as they can stick it on the roof and just run cat5 and low voltage power to the roof. This will give no cable loss and should increase the range yet again, there is also the option of having an external antenna on it, the cable loss would be very minimal due to the antenna being right next to the device. Currently I am unaware of any 802.11g wireless media converters but that shouldn’t be an issue as they will be released soon.

2 Responses to Bristol Wireless Network Report

  1. martian November 26, 2003 at 1:56 am #

    Do you have sample configurations, especially for zebra and hostap? Are the Netgear 311s in ad-hoc mode?

  2. matt December 5, 2003 at 2:08 am #

    The Netgears are in AP mode. There is a nice tarball of pretty much the same setup we use on the sown website aimed at the pebble distro instead of Debian (which we use)

    Links:
    SOWN, http://www.sown.org.uk/
    SOWN pebble meshing, http://www.sown.org.uk/pebble-mesh.tar.gz

    EDIT: We are having a few issues with the me101’s connected directly to a switch (watch this space).