Implementing Atomic Data through Indirect Learning in Dynamic Networks

Abstract

Developing middleware services for dynamic distributed systems, e.g., ad-hoc networks, is a challenging task given that such services deal with dynamically changing membership and asynchronous communication. Algorithms developed for static settings are often not usable in such settings because they rely on (logical) all-to-all node connectivity through routing protocols, which may be unfeasible or prohibitively expensive to implement in highly dynamic settings. This paper explores the indirect learning, via periodic gossip, approach to information dissemination within a dynamic, distributed data service implementing atomic read/write memory service. The indirect learning scheme is used to improve the liveness of the service in the settings with uncertain connectivity. The service is formally proved to guarantee atomicity in all executions. Conditional performance analysis of the new service is presented, where this analysis has the potential of being generalized to other similar dynamic algorithms. Under the assumption that the network is connected, and assuming reasonable timing conditions, the bounds on the duration of read/write operations of the new service are calculated. Finally, the paper proposes a deployment strategy where indirect learning leads to an improvement in communication costs relative to a previous solution that assumes all-to-all connectivity. * This work is supported in part by the NSF Grants 9988304, 0121277, and 0311368. settings is to have the participating network nodes periodically exchange their local state information with the goal of approximating the global state of the system and ensuring progress of local computation [4,8,16]. Performance of a service implemented in this way depends on the prompt update of the local state at each node, hence requiring (logical) all-to-all communication, which can be quite expensive. The communication cost associated with all-to-all communication can be reduced by minimizing the number of bits in the message [2], or by limiting the communication by assigning to each sender a proper subset of the nodes to communicate with [12]. Such methods can lead to good results in static environments, however their utility is diminished in highly dynamic networks. A weakness of all-to-all gossip is its reliance on the existence of point-to-point connectivity. This is an important limitation, since in dynamic systems such as ad-hoc and mobile networks, routing information is prohibitively expensive, where significant amount of power, memory, and communication are needed to keep the routing tables up to date [10,18,19,20]. Furthermore, routing protocols provide a general solution and are oblivious to the data flows of specific applications, which results in unnecessary communication burden. On the other hand, in the absence of a routing service no predictable progress can be ensured in algorithms depending on all-to-all gossip.In this paper we incorporate an indirect learning protocol within a distributed algorithm implementing atomic objects aimed at enhancing it...

Rambo: a robust, reconfigurable atomic memory service for dynamic networks

“…In this section we overview selected results that are either motivated by Rambo or that directly use Rambo as a point of departure for optimizations and practical implementations [9,14,23,24,29,30,37,48,49,58].…”

“…Assuming this may be unfeasible or prohibitively expensive to implement in dynamic networks, such as ad hoc mobile settings. Konwar et al [37] develop an approach to implementing Rambo service where all-to-all gossip is replaced by an indirect learning protocol for information dissemination. The indirect learning scheme is used to improve the liveness of the service in the settings with uncertain connectivity.…”

“…6, lines [22][23][24][25][26][27][28][29][30][31][32][33][34][35][36][37][38][39], the first step it takes is to set its status to active, if it has not already done so (line 25). It then adds any new information about the world contained in the message to its local world set (line 26).…”

“…This can in fact happen when node i gets too far out-of-date, i.e., all messages to and from i are lost for a longer period of time, during which multiple configurations are installed and removed. When this happens, the processing of any ongoing read/write operation must be restarted, beginning again the phase in progress (lines [35][36][37]. (The concept of restarting is only with respect to internal progress while processing of the read/write operations, and has no affect on the standard atomic semantics of read/write operations: from a client's perspective, the restart is invisible.)…”

“…(The concept of restarting is only with respect to internal progress while processing of the read/write operations, and has no affect on the standard atomic semantics of read/write operations: from a client's perspective, the restart is invisible.) When the configuration map becomes unusable, the process of accessing quorums is started again (lines [35][36][37]: the local phase number is incremented, the set op.acc is reset to ∅, and the op.cmap is reset to cmap, the most recently known configuration map. Otherwise, if there is no need for a restart, then the sender of the message is added to op.acc (line 33).…”