Overlapping Cells:

Placement of the access points is important to ensure proper coverage and can also be used to increase bandwidth and alleviate the load on an access point.  Each access point provides coverage of an area called a cell.  To avoid dead spots between cells, plan for
sufficient overlap between adjacent cells.  For dead spots at the edge for periphery cells, make sure the  outer edge of the cell extends sufficiently far beyond the edge of the
classroom, otherwise transient interference may suddenly disconnect students on an edge of the classroom.  Laptop-to-access point coverage is obviously important, but interactive wireless classrooms also require laptop-to-laptop coverage. Although less likely, it is possible that two laptops in the same large room use the same access point, but cannot communicate with one another.  In this case, adding additional (carefully placed) access points can solve the problem.

Overlapping cells can also be used to increase the access point to laptop ratio, thereby reducing loads on an access point and increasing the aggregate network bandwidth supported.

Finally, care must be taken to avoid routing loops caused by overlapping access points communicating with one another.  In general, most access points are designed to prevent such loops and often support spanning tree algorithms to avoid them, but it is not
difficult to accidentally create loops that bring down the wired network as well.

Current wireless LAN technology operate at rates of roughly 2 Mbps while conventional ethernets  operate at 10 or 100 Mbps.  Although wireless speeds are expected to increase to 10 Mbps before year's end, they are not likely to match leading-edge wired speeds any
time soon.  This speed mismatch can be a problem.  There is no way that all the packets on the wired network can be forwarded to the wireless network.  To solve this problem, most access points have the ability to filter traffic based on destination or traffic class.
Thus traffic not originating from or destined to wireless machines never appears on the wireless network.  However, multicast and broadcast traffic can still be a problem because they is more difficult to filter.  Although one can argue that networks with substantial broadcast/multicast traffic are poorly designed or use poorly designed applications, the fact is that such networks are quite common.  The College of Engineering network to
which we connected is one example (for a variety of reasons) and so broadcast traffic was initially consuming a significant portion of our wireless network's 2 Mbps bandwidth.
Fortunately the access point allowed us to filter the majority of the broadcast traffic by tossing packet types which we were not using.  Note that disabling multicast/broadcast is an option but protocols like DHCP use broadcasting to assign IP addresses.  Our access point also allowed us to set a maximum number of broadcast/multicast messages allowed per second which is a useful feature.
 

When designing the network, it is important to decide what is required: mobility or roaming. We define mobility to be the act of stopping all network communication, moving to a new location serviced by a different access point, and restarting all network connections.  Roaming removes the requirement of stopping all network connections.  A roaming machine can move from one access point to another while all network connections and data transfers continue on uninterrupted.   Wireless classroom environments do not require roaming support since it is usually acceptable for a student to be ``off'' the network while moving between classes.

While mobility is possible with most vendors and operating systems, roaming is not.
Roaming requires that access points cooperate to seemlessly ``handoff'' a machine as it moves from one cell to another.  This typically requires handoff support in the access points and roaming support in the operating system.  For example, we could achieve
mobility in the version of Linux we were using, but not roaming, while both were supported under Windows 95.  Recently we found card services management software for Linux that does support roaming (see  RASCAL Project ).  Note that roaming typically assumes continuous coverage from the starting cell to the destination cell.
 

Even if mobility and roaming are supported, there are limitations that you should consider.  First, if the access points are connected to different IP networks, roaming is probably not supported (there is research going on, but you are not likely to find
support for this in off-the-shelf hardware.).  Second, if the access points are attached to a bridged network, roaming may or may not be supported.  If ``transparent'' bridging is
used, roaming will work as expected.  However, some learning or smart bridging hardware may become confused when a laptop moves from one access point to another and will incorrectly deliver its messages.

We have three buildings covered by our wireless network.  The networks in each of these buildings are separated by one or more switches and/or routers.  Therefore, we cannot roam from building to building.  Whenever we move from one building to another building, we must re-establish the network connection and in some cases use DHCP to get a new network address.  This is not a problem for us since re-establishing a network connection on a laptop using PCMCIA card/antennas is simply a matter of pulling out the PCMCIA card then putting it back into the slot.