Adding a node to the testbed
Deploying a node is a 2-stage procedure: (a) installing a node into the testbed and (b) placing the node in a specific location. The second stage is described under "Deployment techniques" below.
Node installation
The following steps describe exactly how to install a node into the testbed. It is good to have an Ethernet port available nearby, which should be already connected to one of the testbed's PoEswitches. Otherwise, you need to have a small PoE switch next to the server, for testing purposes. For the steps below, we assume that you have an available Ethernet port connected to one of the PoE switches. Note that the following steps can be used for any Soekris net_x series product. We describe the procedure for the net4826 that we are using.
1. Assemble the node, by installing the WiFi card, the pigtail and the external antenna. Carefully connect the pigtail to the WiFi card. Screw the case on the Soekris box, and connect the D-Link-P50 both on the Ethernet interface and the power input. Connect the D-Link-P50 to the testbed switch, via the nearby PoE Ethernet port, and enable PoE on this port. You should see the leds of the Soekris box blinking, which implies that the node is powered-on.
2. Remove the PoE supply, connect an appropriate external power adapter and then connect the node's serial port to the server's serial port. For this, you will need a DB9 female-to-female RS232 cable. Run the "minicom" application at the server. (Install the minicom application if it is not there already).
3. Select a baud rate of 19200, by using CTRL+A+P, F, and reboot the node. At its nominal start the node will start counting its memory. After the counting is done, the node will identify its Ethernet address and it will send a bootp request to the server. At this point, nothing else will happen. This is because the server will receive the request, but it will not identify the MAC address of the node.
4. Write down the MAC address of the node, which should appear on the server's monitor, from the minicom application. Then, edit the /etc/bootptab file of the server, and add an entry for the new node. (We have explained how to do this, in the Client set-up procedure, in the Software section). Make sure that you enter the MAC address correctly, and that you assign a unique IP address to this new node. Save the bootptab file, and reboot the node.
5. The node will now boot the linux kernel from the server and will load the kernel and all the appropriate modules. After that, the node is accessible remotely. We can ssh the node through its IP address, via the server.
6. Power the node off, and go to the location wherein the node will be placed. After proper negotiations with the people around, place the node and connect the D-Link-P50 to the new Ethernet port at that location.
7. Connect the selected Ethernet port to the switch. After this, the connection should be in the way depicted in the figure. Enable the PoE capability on the new switch port. The node should now work fine, and should be accessible remotely from the server as well. Note that it takes between 2-4 minutes for the node to finish booting.
Deployment techniques
There are different techniques that can be followed for node deployment.
First, one could place nodes randomly, regardless of the environmental variability. This approach however is quite risky. Even if the WiFi card is very powerful, we may not always want nodes to transmit with very high power. As a result, with lower power, a randomly placed node may lose connectivity from the rest of the network.
As another method , one could take into account the variability and the obstacles around the potential location. For example, one may avoid placing a node near a door that people open and close frequently. In addition, microwave ovens and portable phones (2.4 GHz band) may also cause significant interference to nearby receivers. Even though the environmental variability is unpredictable to a large extent, with this approach one should try to reduce its effects as much as possible, by placing nodes away from such devices.
It is good to have a map of each floor of the builing, together with the locations of the previously deployed nodes. This way one has an idea of the achieved topology after placing the new node. Our current map, from the 3rd floor, is given below. Always, keep in mind how powerful is the wireless card that you are using. Materials, such as glass, bricks and metallic objects create very unpredictable connectivities.
Problems and tricks
After experimenting for quite some time, we noticed the following:
- Using a desktop PC as a node: This is not such a good idea. Besides the fact that desktop PCs become noisy after some time, they appear to be inefficient when two or more WiFi cards are plugged-in. We tested this using a trivial desktop PC, in which we installed two PCMCIA cards. When the cards are transmitting at the same time, the achieved throughput is the half of what the cards can provide in isolation.
- If we put two or more nodes next to each other inside the same room, they can't communicate efficiently with high power. We experimented with different cards and different frequencies.
- It is better not to use the maximum supported transmission power for the cards. This is not the fact for all cards, however, it is for the one that we are using. This probably has to do with hardware deficiencies of the WiFi card.
- The connectivity is different for different frequency channels, as one may expect.
