Proactive replication strategies, which generally require
positioning systems on client devices and
significant processing resources for predicting node disconnections, are unsuitable for dense MANETs.
REDMAN proposes an original lightweight RDM facility that works to maintain unchanged the replication degree
decided for each shared resource, without guaranteeing absolute consistency, i.e., it is possible to have
time intervals when the requested replication degree differs from the actual number of replicas in the dense MANET.
In other words, RDM only aims at maintaining consistency likelihood for the chosen resource replication degree,
with the relevant pro of significantly limiting the overhead on participant nodes.
The replica manager is in charge of enforcing the needed replication degree and of maintaining information
about all shared resources available in the dense MANET (Shared Resource Table - SRT).
For any shared resource, the corresponding SRT entry includes the associated replication degree and
weakly consistent information about the nodes where the resource is currently replicated.
When a delegate enters a dense MANET, it communicates the metadata of the shared resources to the replica manager,
which decides the replication degree to enforce by considering the hint about resource criticality
and the number of current participants in the dense MANET.
Future work will improve the replica manager decision capability, by taking into account additional indicators
such as the current and average network load. Afterwards, the manager creates new SRT entries for the newly arrived
resources; any entry contains the enforced replication degree and weakly consistent information about the nodes
currently hosting replicas for that resource. Finally, the manager commands the delegates to start the replication
operations.
After the initial replica distribution performed by RD, RDM simply works reacting when resource delegates leave the
dense MANET.
Three different cases may occurr:
If a delegate can realize it is going to exit from the dense region (see the following section)
it autonomously replicates its shared resources on adjacent nodes belonging to the dense MANET;
these nodes then notify the replica manager of the occurred change.
In Figure 1, node D realizes it is going to leave the dense MANET when it still belongs to the network.
Thus, it still has enough time to offload its shared resources to node E. Node E, upon accepting to replicate the
resources, notifies the manager that consequently updates SRT.
Figure 1: Node D realizes it is going to leave the network.
If a delegate is not able to foresee
its leaving in time, once it realizes not to be in the dense region anymore, it tries to advise the manager
(probably still reachable via intermediate nodes near the dense MANET boundaries) by specifying its hosted
shared resources; it is the replica manager that commands to other delegates for those resources, as indicated in SRT,
to distribute new replicas so to re-establish the required replication degrees.
In Figure 2, D realizes it has left the networks only when it is out of dense MANET. By exploiting a possibly
unstable connection it advises the manager of its departure. The manager updates SRT (i.e., it deletes D entry from
all D shared resource entries) and commands a new replica diffusion to node L (that is present in the entry's Probable
Replica Placement).
Figure 2: Node D realizes it has left the network when an
unstable connection is still available.
The above strategy does not guarantee strict consistency at any time between the actually maintained and
the requested replication degree, e.g., because of message losses or abrupt failures/exits of delegates from
the dense region. For this reason, at large time intervals (AUM Periods),
resource delegates periodically send to the manager the list of their locally hosted shared resources (AUM message);
in the case that the manager relizes (by checking AUM messages every
AUM Check Periods) a delegate stops sending update messages,
the manager commands new replicas of its resources to other delegates and correspondingly updates SRT.
In Figure 3, node D fails (or equivalently abandons abruptly the dense MANET). The manager realizes the problem only
when it checks AUM messages consistency (it receives fresh updates from L and A, but not from D).
Then, it updates its SRT (by deleting D entry) and commands a new replication to node L.
Figure 3: Node D fails or abruptly leaves the network.
When performs AUM check, the manager realizes the abandon.