Ordered Message Delivery Schemes for Group Communication in Distributed and Mobile Computing Systems

• Main Authors: Chih-Ming Hsiao, 蕭志明
• Other Authors: Ge-Ming Chiu
• Format: Others
• Language: en_US
• Published: 2005
• Online Access: http://ndltd.ncl.edu.tw/handle/86917065036937794273

博士 === 國立臺灣科技大學 === 電機工程系 === 93 === In distributed computing systems, processes are often organized into groups for supporting various applications. Ordered delivery of messages in group communication has been an important issue. Sequence-number-based mechanisms have proven to be a simple, efficient, and practical approach to achieve total ordering delivery. There are two challenging issues that are faced in implementing sequence-number-based scheme for total ordering requirement. The first one is fixed sequencer method to generate the sequence number has poor fault tolerance as the sequencer may become a single point of failure. The other one is the sequencer site may become a bottleneck in terms of performance as all requests are addressed to the single site and the single sequencer does not scale to large group size or multiple groups. In this dissertation we first adopt a sequencer-based scheme to large group size or multiple groups. We introduce the concept of coordinating sequencers that arms to constructs a sequence array for a message to achieve the total ordering requirements for group communication systems with multiple overlapping groups. Our scheme assigns a sequence to each group. For a given message, the sequencer of the destination group constructs a sequence array by requesting for relative delivery positions from the sequencers of the overlapping groups. The sequence array is then used by any receiving process to determine the delivery sequence of the message. A salient feature of the protocol is that the coordination between the sequencers is performed in a simple, asynchronous and non-blocking manner. The delivery operation at a receiving process is very simple, and a message can be delivered as soon as it becomes deliverable. Second, we address the messages ordering requirement in mobile computing system. There is a growing trend in developing applications for mobile computing systems in which mobile host computers retain their network connections while in transit. Mobile devices typically have severe constraints in terms of energy, computing and storage resources. Wireless channels used by mobile devices have significantly lower bandwidth than fixed communication links. Henceforth, it is important that group communication protocols designed for mobile computing environments keep the computing load and the communication overhead on mobile hosts low. Smooth host migration mechanism is also demanded for such protocols. In this dissertation we propose an efficient adaptation of the propagation-tree technique to mobile computing systems with multiple overlapping groups. It takes advantages of the capabilities of stationary mobile support stations to overcome the deficiencies associated with mobile devices. We construct the propagation tree based on the stationary stations, rather than the mobile hosts. As a result, mobile hosts are relieved of the excessive load of forwarding messages and communications on wireless channels are confined to transmitting messages to destination processes. Moreover, the proposed protocol employs a mechanism to synchronize transmissions within a wireless cell. This serves to avoid redundant transmissions of a message in a wireless network in an attempt to achieve better utilization of the network bandwidth. Our mechanism relies on a handoff operation to deal with mobility of mobile devices. The handoff procedure ensures a smooth integration of a mobile host into a new cell, while preserving reliability of communication and total ordering of message delivery. Next, we address a fault-tolerant scheme for generating global sequence numbers for total ordering requirement in group communication. Our solution is to maintain the information about the latest-used sequence number at multiple participant nodes for enhancing fault tolerance. In the protocol, each process may initiate the generation of sequence numbers independently for messages emitted by itself. No single process failure may crash the operation of the system. Final, we also introduce a two-tier causal ordering protocol for wireless access network which support multihop wireless infrastructures. Our protocol is essentially an adaptation of ISIS CBCAST technique to mobile computing systems with multihop wireless infrastructures. The size of vector time on message is proportional to the number of mobile hosts in the zone. In addition, the delivery operation at a receiving mobile host is very simple, and a message can be delivered as soon as it becomes deliverable. These factors may amount to a significant saving of computing and communication overhead for mobile computing systems in which mobile devices are typically tight on resources. Furthermore, our mechanism relies on a handoff operation to deal with mobility of mobile devices. Cell switchings of mobile hosts do not trigger any message exchange in the wired network. The handoff procedure ensures a smooth integration of a mobile host into a new zone, while preserving the causal ordering property of message delivery.