Mali

Design choices

The technical team determined that the access point would be installed at 20 meters up the radio station tower, just below the FM radio dipoles, and not so high as to interfere with coverage to client sites below in the bowl-like depression where most were found. The team then focused on how to connect each client site to this site. An 8 dBi omni (from Hyperlinktech, hyperlinktech.com) would suffice, providing coverage to all client sites. The 8 dBi antenna that was chosen has a 15 degree down-tilt, assuring that the two clients less than a kilometer away could still receive a strong signal. Some antennae have very narrow beam width and thus "overshoot" sites that are close. Panel antennae were considered, though at least two would be required and either a second radio or a channel splitter. It was deemed unnecessary for this installation. The following calculation shows how to calculate the angle between the client site's antenna and the base station's antenna, using standard trigonometry.

tan(x) = difference in elevation

+ height of base station antenna

-height of CPE antenna / distance between the sites

tan(x) = 5m + 20m - 3m / 400m x = tan-1 (22m / 400m) x =~ 3 degrees

In addition to the equipment in the telecentre (4 computers, a laser printer, 16 port switch), the radio station itself has one Linux workstation installed by the

author's project for audio editing. A small switch was installed in the radio station, an Ethernet cable was run through plastic tubing buried at 5 cm across to the telecentre, across the yard.

From the main switch, two cables run up to a Mikrotik RB220, access point. The RB220 has two Ethernet ports, one that connects to the VSAT through a cross-over cable, and the second that connects to the radio station's central switch. The RB 220 is housed in a D-I-Y PVC enclosure and an 8 dBi omni (Hyperlink Technologies) is mounted directly to the top of the PVC cap.

The RB220, runs a derivative of Linux, Mikrotik version 2.8.27. It controls the network and provides DHCP, firewall, DNS-caching and routes traffictothe VSAT, using NAT. The Mikrotik comes with a powerful command line and a relatively friendly and comprehensive graphical interface. It is a small x86 based computer, that is designed for use as an access point or embedded computer. These access points are POE capable, have two Ethernet ports, a mini-pci port, two PCMCIA slots, a CF reader (which is used for its NVRAM), are temperature tolerant and support a variety of x86 operating systems. Despite that the Mikrotik software requires licensing, there was already a substantial install base in Mali and the system has a powerful and friendly graphical interface that was superior to other products. Due to the above factors the team agreed to use these systems, including the Mikrotik software to control these networks. The total cost of the RB220, with License Level 5, Altheros mini-pci a/b/g and POE was $461. You can find these parts at Mikrotik online at www.mikrotik.com.

The network was designed to accommodate expansion by segregating the various sub-networks of each client; 24 bit private subnets were alloted. The AP has a virtual interface on each subnet and does all routing between, also allowing fire-walling at the IP layer. Note: this does not provide a firewall at the network layer, thus, using a network sniffer like tcpdump one can see all traffic on the wireless link.

To limit access to subscribers, the network uses MAC level access control. There was little perceived security risk to the network. For this first phase, a more thorough security system was left to be implemented in the future,, when time could be found to find an easier interface for controlling access. Users were encouraged to use secure protocols, such as https, pops, imaps etc.

The affiliate project had installed a C-band VSAT (DVB-S) system. These satellite systems are normally very reliable and are often used by ISPs. It is a large unit, in this case the dish was 2.2 meters in diameter and expensive, costing approximately $12,000 including installation. It is also expensive to operate. A 128 kbps down and 64 kbps up Internet connection costs approximately $700 per month. This system has several advantages compared to a Ku system though, including: greater resilience to bad weather, lower contention rates (number of competing users on the same service) and it is more efficient at transferring data.

The installation of this VSAT was not ideal. Since the system ran Windows, users were able to quickly change a few settings, including adding a password to the default account. The system had no UPS or battery back up, so once a power outage occurred the system would reboot and sit waiting for a password, which had since been forgotten. To make this situation worse, because the VSAT software was not configured as an automatic background service it did not automatically launch and establish the link. Though the C-band systems are typically reliable, this installation caused needless outages which could have been resolved with the use of a UPS, proper configuration of the VSAT software as a service, and by limiting physical access to the modem. Like all owners of new equipment, the radio station wanted to display it, hence it was not hidden from view. Preferably a space with glass doors would have kept the unit secure while keeping it visible.

The wireless system was fairly simple. All of the client sites selected were within 2 km of the radio station. Each site had a part of the building that could physically see the radio station. At the client site, the team chose to use commercial, client grade CPEs: Based on price, the Powernoc 802.11b CPE bridge, small SuperPass 7 dBi patch antennas and home-made Power Over Ethernet (POE) adaptors. To facilitate installation, the CPE and the patch antenna were mounted on a small piece of wood that could be installed on the outside wall of the building facing the radio station.

In some cases the piece of wood was an angled block to optimize the position of the antenna. Inside, a POE made from a repurposed television signal amplifier (12V) was used to power the units. At the client sites there were not local networks, so the team also had to install cable and hubs to provide Internet for each computer. In some cases it was necessary to install Ethernet adapters and their drivers (this was not determined during the assessment). It was decided that because the client's networks were simple, that it would be easiest to bridge their networks. Should it be required, the IP architecture could allow future partitioning and the CPE equipment supported STA mode. We used a PowerNOC CPE bridge that cost $249 (available at powernoc.us).

Local staff were involved during the installation of the wireless network. They learned everything from wiring to antenna placement. An intensive training program followed the installation. It lasted several weeks, and was meant to teach the staff the day to day tasks, as well as basic network troubleshooting.

A young university graduate who had returned to the community was chosen to support the system, except for the cable installation, which the radio station technician quickly learned. Wiring Ethernet networks is very similar to coaxial cable repairs and installations which the radio technician already performed regularly. The young graduate also required little training. The team spent most of its time helping him learn how to support the basics of the system and the telecentre. Soon after the telecentre opened, students were lined up for the computer training, which offered 20 hours of training and Internet use per month for only $40, a bargain compared to the $2 an hour for Internet access. Providing this training was a significant revenue and was a task that the young computer savvy graduate was well suited for.

Unfortunately, and somewhat unsurprisingly, the young graduate left for the capital, Bamako, after receiving an offer for a government job. This left the telecentre effectively marooned. Their most technically savvy member, and the only one who was trained in how to support the system, had left. Most of the knowledge needed to operate the telecentre and network left with him. After much deliberation, the team determined that it was best not to train another tech savvy youth, but rather to focus on the permanent local staff, despite their limited technical experience. This took much more time. Our trainers have had to return for a total of 150 hours of training. Several people were taught each function, and the telecentre support tasks were divided among the staff.

Training did not stop there. Once the community services were connected, they too needed access. It seemed that although they were participating, the principals, including the mayor, were not using the systems themselves. The team realized the importance of assuring that the decision makers used the system, and provided training for them and their staff. This did remove some of the mystique of the network and got the city's decision makers involved.

Following training, the program monitored the site and began to provide input, evaluating ways that this model could be improved. Lessons learned here were applied to other sites.

Financial Model

The community telecentre was already established as a non-profit, and was mandated to be self-sustaining through the sale of its services. The wireless system was included as a supplementary source of revenue because early financial projections for the telecentre indicated that they would fall short of paying for the VSAT connection.

Based on the survey, and in consultation with the radio station whom manages the telecentre, several clients were selected. The radio station negotiated contracts with some support from its funding partner. For this first phase, clients were selected based on ease of installation and expressed ability to pay. Clients were asked to pay a subscription fee, as described later.

Deciding how much to charge was a major activity which required consultation and expertise that the community did not have in financial projections. The equipment was paid for by the grant, to help offset the costs to the community, but clients were still required to pay a subscription fee, which served to assure their commitment. This was equivalent to one month of the service fee.

To determine the monthly cost for an equal slice of bandwidth we started with the following formula:

VSAT + salaries + expenses (electricity, supplies) = telecentre revenue + wireless client revenue

We had estimated that the telecentre should earn about $200 to $300 per month in revenue. Total expenses were estimated to be $1050 per month, and were broken down as: $700 for the VSAT, $100 for salaries, $150 for electricity, and about $100 for supplies. About $750 in revenue from the wireless clients was required to balance this equation. This amounted to roughly $150 from each client. This was just tolerable by the clients, and looked feasible, but required fair weather, and had no room for complications.

Because this was becoming complicated, we brought in business geeks, who modified the formula as such:

Monthly expenses + amortization + safety funds = total revenue

The business experts were quick to point out the need of amortization of the equipment, or one could say "re-investment funds" as well as safety funds, to assure that the network can continue if a client defaults, or if some equipment breaks. This added about $150 per month for amortization (equipment valued at about $3,000, amortized over 24 months) and the value of one client for default payments, at $100. Add another 10% to account for currency devaluation ($80), and that equals an expense of $1380 per month. In trying to implement this model, it was finally determined that amortization is a concept that was too difficult to convey to the community, and that they would not consider that clients might default on payment. Thus, both formulae were used, the first by the telecentre and the second for our internal analysis.

As was soon discovered, regular payments are not part of the culture in rural Mali. In an agrarian society everything is seasonal, and so too is income. This means that the community's income fluctuates wildly. Moreover, as many public institutions were involved, they had long budget cycles with little flexibility. Although they theoretically had the budget to pay for their service, it would take many months for the payments to be made. Other fiscal complications arose as well. For example, the mayor signed on and used the back-taxes owed by the radio to pay for its subscription. This of course did not contribute to cash flow. Unfortunately, the VSAT providers have little flexibility or patience, as they have limited bandwidth and only have room for those that can pay.

Cash flow management became a primary concern. First, the revenue foreseen in financial projections showed that even with an optimistic outlook, they would not only have trouble earning enough revenue on time to pay the fee, but getting the money to the Bamako-based bank also presented a problem. Roads near the village can be dangerous, due to the number of smugglers from Guinea and wayward rebels from the Ivory Coast. As projected, the telecentre was not able to pay for its service and its service was suspended, thereby suspending payment from their clients as well.

Before the project was able to find solutions to these problems, the cost of the VSAT already began to dig the telecentre into debt. After several months, due to technical problems, as well as concerns raised in this analysis, the large C-band VSAT was replaced with a cheaper Ku band system. Although cheaper, it still sufficed for the size of the network. This system was only $450, which by ignoring amortization and safety margins is affordable by the network. Unfortunately, due to default payments, the network was not able to pay for the VSAT connection after the initial subsidized period.

Conclusions

Building a wireless network is relatively easy, but making it work is much more of a business problem than a technical problem. A payment model that considers re-investment and risk is a necessity, or eventually the network will fail. In this case, the payment model was not appropriate as it did not conform to fiscal cycles of the clients, nor did it conform to social expectations. A proper risk analysis would have concluded that a $700 (or even a $450) monthly payment left too narrow a margin between revenue and expenses to compensate for fiscal shortcomings. High demand and education needs limited the expansion of the network.

Following training the network operated for 8 months without significant technical problems. Then, a major power surge caused by a lightning strike destroyed much of the equipment at the station, including the access point and VSAT. As a result, the telecentre was still off-line at the time that this book was written. By that time this formula was finally deemed an unsuitable solution.

'Ian Howard