The untraditional routing algorithms, deal with different quality metrics associated with nodes, location or other information used to choose a suitable relay node. On the other hand, most traditional routing policies deal with only address and interface metrics, where the objective is the management of network traffic in order to decrease its latency, cost, congestion and other metrics. For example, policies for BGP, the current inter-domain routing protocol in the Internet, may be used to update routing tables and consequently to control the traffic forwarding, taking into account path lengths, local preference values, traffic origin among others factors. In terms of software, the IP Filter, iproute2 and net filter are utilities commonly used to create and apply network policies in order to control the routing in local area networks.
However, these approaches are based on the origin-destination pair of IP addresses and interfaces and are not suitable for 4G networks where routing involves often multi-homed end-hosts, intermittent connection, unpredicted mobility and other dynamic characteristics. We need new policies and mechanisms to deal with buffer, message replication and scheduling schemes.
In untraditional routing algorithms, information forwarding can be based on replication (usually broadcast) or on the basis of the pair: source and destination nodes. When using replication, one needs to evaluate and establish the necessary number of replicas according to the underlying environment being used (network size, mobility model, etc.). This is important to avoid network congestion and unnecessary overhead and resource usage. Note that a replication level or value must be carefully set as a small number of replicas may very well become insufficient for the dissemination of information and discovering all nodes and paths. Changes to the number of neighbors over a short period of time indicate a high mobility scenario and consequent frequent topology updates including the network size. This information is a very determinant factor in setting the replication policy and its parameters.
One may choose to apply a policy over the forwarding mechanism based on source and destination information, whereby a node is not required to send all messages. Forwarding may be subject to priority in order to give preferential treatment to some messages for example or offer differentiated traffic. Work in (Lindgren & Phanse, 2006) evaluated four such strategies: GRTR, GR-TRSort, GRTRMax and COIN. GRTR defines a mechanism that allows forwarding to a neighbor if the encountered node has higher delivery predictability to the message destination. When a node A uses GRTR Sort with its neighboring node B, node A subtracts both delivery predictability values for this message destination and then forwards the message for node B using the highest subtraction result it could find. The GRTR Max strategy is also a little similar to GRTR and the difference between these lies in the ordering executed by GRTRMax which is before comparing the delivery predictability. The last policy example is given by COIN which generates a random value for each message. Depending on whether this is higher than 0.5, then the node forwards the message which is similar to tossing a coin and taking a decision according to the outcome.
Independently of whether the message forwarding policy is based on broadcast, replication use, or according to source and destination, the buffer (where messages are stored before being forwarded) is obviously not an infinite resource. Therefore, one needs to have one or more policies in place to determine what to do with the messages. Common per hop behavior or queue management techniques such as FIFO (First In First Out), MOFO (Most Forwarded first) – messages, MOPR (Most favorably forwarded first), SHLI (Shortest Life Time First) or LEPR (Least Probable First) may apply.
Policies may also determine for how long a message could be stored. For example, the Data Mule project considers that a mobile node must store information until finding or coming into contact with a destination (in this case, a base station within a sensor network). Moreover, Data Mule chooses the next hop according to the mobile's behavior, all the way from the fixed nodes, by forwarding the data to intermediary mobile nodes that collect and store the information until the base station is reached.
Another routing work involving policies. In this social overlay shaped by groups, a node needs to decide which query could be dropped, when the buffer is full or simply choose randomly queries to swap. However the authors suggest that the nodes should firstly exchange meta-information, before deleting and adding queries (buffer management). Although it uses a simple decision mechanism, it follows the policy approach to choose what queries must be dropped from the buffer and exchanged with its neighbors. Instead of using such a simple policy model, it could adopt a social policy that gives higher priority to messages forwarded from friends or partners over strangers or unknown senders.
In summary, this section presented how important it is to work towards reducing costs and the requirement for policy changes according to behavior modification in order to provide more flexibility for routing mechanisms. Moreover, it has shown that policies may be applied to control several device features (i.e. buffer, message replication and message scheduling schemes) to determine, configure and improve 4G routing.
No comments:
Post a Comment