We have performed both in-the-field and simulation experiments to characterize maximum throughput in relation to path length and active clients.
In the former case our experiments clearly show that competing devices tend to fairly share the totally available bandwidth (Fig. 1 - left). The standard deviation, that grows while increasing the number of clients, is mainly due to very few and occasional short time intervals in which a single client tends to get greater throughput than the others. Those experimental results in saturation conditions justify our conservative simplifying assumption about throughput: in a first approximation, we consider the maximum throughput achievable by an MMHC node as inversely proportional to the number of active nodes on that single-hop link.
In addition, we have experimentally observed that throughput tends to degrade with the number of traversed hops (Fig. 1 - right). For a coarse-grained lightweight approximation, we can adopt an average per-hop degradation of 20% to the purpose of MMHC management decisions. These practical performance aspects suggest considering only rather flat MMHC topologies, with few hops and a limited number of clients, especially for Bluetooth connectivity.
Let us rapidly note that the reported results have been measured by using heterogeneous wireless interfaces by different manufacturers, e.g., Orinoco Gold, Buffalo, and PRO/Wireless for IEEE 802.11, Mopogo dongles, and Centrino integrated hardware for Bluetooth v1.2. We have experienced the same trends but non-negligibly better performance in deployment environments with homogeneous hardware. In the following, we always report results from heterogeneous scenarios (worst case), also because they better mimic the open environments for self-organizing network connectivity we intend to address.
Fig. 1. MMHC throughput exhibits an almost linear degradation with the number of clients (left) and of traversed hops (right) for both IEEE 802.11 (up) and Bluetooth (down).
In addition to in-the-field experiments, we have thoroughly evaluated multi-hop multi-path performance also in a simulated environment, as needed to assess the proposed solutions in articulated scenarios with a large number of mobile nodes.
Fig. 2 - left shows the simulation-based average throughput of a multi-hop path depending on its length, in saturated bandwidth conditions. The one-hop path supports throughput slightly greater than 5Mb/s (around the typical maximum throughput of IEEE 802.11b). By increasing the length of the sender-receiver path, the throughput lowers dramatically (less than 1Mb/s in a 7-hop path). This is mainly due to interference of neighbor nodes, especially relevant when different wireless interfaces on the same node exploit similar frequencies. In fact, each intermediate node usually serves as a forwarder, thus potentially producing collisions among incoming and outgoing packets. This effect increases with longer paths (larger number of intermediate nodes).
Fig. 2 - right shows the simulated average throughput depending on the number of senders, each one at single-hop distance from the receiver. Senders have shown to share the available bandwidth with sufficient fairness: only with 5 or more senders the normalized throughput (i.e., the ratio between the standard deviation and the average value of throughput) slightly increases. In other words, to a first approximation, even when there are many senders, generally the throughput is fairly shared and almost constant (slightly greater than 5Mb/s).
Fig. 2. Left: single-client throughput at increasing path length. Right: single-hop throughput at increasing number of clients: average (line), standard deviation (error bars), normalized throughput (impulses).