Using structural characteristics for autonomous operation

• Authors: Carlos Baquero;Francisco Moura
• Venue: ACM SIGOPS Operating Systems Review
• Year: 1999

The majority of current mobile computing systems operate either in conjunction with a central network by some form of weak connectivity or tend to operate in total isolation and perform sporadic synchronization with a backup or a central network. These configurations miss an additional and very useful pattern of operation --- mobile to mobile interaction. Recent mobile devices have the capacity for direct communication among them, but this option is essentially neglected by the application software.In order to address this pattern of operation we believe that there is a need to support re-usable peer-to-peer synchronization mechanisms that both respects data ownership and enables some level of state reconciliation.Naming this operation pattern as autonomous operation, we can observe that this pattern is already found on many legacy applications deployed in distributed systems. For example, personal information managers, Mail/News readers and Web browsers, often store persistent state in local files, but tacitly assume a single copy. Noticing that these separate copies are in fact replicas of a distributed entity, leads to the creation of semantically knowledgeable file synchronizers that strive to restore an unified state from these replicas.Evolution from static distributed systems to mobile platforms raises a demand for applications that, not only are adapted to user mobility but, take advantage of it. It is clear that despite continuous improvements on connectivity support for mobile environments, the cost and coverage limits still imply a major share of disconnected operation. When connectivity does exist it usually interposes wide area networks between communication peers, when one party is on the road, leading to lower channel quality. On the other hand, user mobility is likely to conduce to, normally unforeseen, physical proximity of the user's mobile computer with other mobile or fixed systems. This occurrence is likely to increase as the installed population of mobile devices increases.In this work we show that without imposing restrictions on availability, which is a crucial factor for personal applications, it is possible to enable some data sharing among autonomous mobile applications. This sharing would take advantage of any pairwise encounters of replica holders.To determine the level of sharing that is compatible with permanent availability, we model general purpose data types that provide the necessary reconciliation guarantees. These guarantees are obtained by placing restrictions on the allowed behavior in order to avoid the occurrence of conflicting concurrent operations that would prevent reconciliations. Among other uses, these data types should help to identify sharable segments of data on classes of applications that traditionally support no sharing at all, and identify which parts of the state can be effectively shared.In the next section we present some examples of sharable data that motivates the modeling of a more generic and higher level description. This description is presented in the third section together with a framework of convergent components. Section four builds on this framework and gives a general presentation of a Java implementation for a component hierarchy. Before presenting the conclusions we show how these tools where used to build a merger for pairs of bookmark files, giving some insight on how to combine the components to create concrete applications.