Self-Adaptive Clock Synchronization Based Upon Characteristics Of Asynchronous Network Environment

The problem of Clock Synchronization is a classical issue in distributed operating system and core technology in distributed computing. As the development of computing technologies and network technologies rapidly, the new content and broad range are added in this research area of clock synchronization. The continuous time-stamp communication model was applied to evaluate running precision difference of clocks in order to reduce the effect of network delay efficiently. In order to synchronize physical clocks in asynchronous network environments, the thesis applied leader election algorithm to simplify architecture of clock synchronization system. Designing clock synchronization algorithms presents a number of difficulties. First, due to variations of transmission delay each computer cannot have an instantaneous time value to synchronize clocks. Second, even if all clocks could be started at the same real time, they would not remain synchronized because of drifting rates. In fact, clocks run at a rate that can differ from real time by 10-6 seconds per second and thus can drift apart by near one second per 11 days. In addition, their drift rate can be changed due to temperature variations or aging. The difference between two hardware clocks can thus change as time passes. Finally, the most important difficulty in the research of clock synchronization is how to support faulty and self-adaptive capabilities of system. The research of this dissertation used continuous timed-stamp communication model to create statistical mathematical models of clock running precision difference. The application of continuous time intervals can achieve better precision than that of instantaneous time intervals,because this model was created by a lot of time-stamps accepted by this node. The error or drop of a little time stamps could not affect the precision of this model. The performance is evaluated timely and dynamically by clock synchronization graph based on clock precision differences. The stability of clock synchronization can be determined by exchanging messages of clock precision difference. There are two kinds of clock synchronization tasks: external synchronization and internal synchronization. The key difference of these two implementation methods is the presence of a stand time basis. If a time basis exists, we usually use the external synchronization policy to synchronization all clocks; otherwise the internal clock synchronizationmethod will be applied. Because there is not a standard time coordinate in internal clock synchronization system , this issue becomes more complicated. In order to simplify the design and implementation of internal clock synchronization system, a leader election mechanism for internal clock synchronization is introduced, where the leader node will be considered a reference node, and views internal clock synchronization as external clock synchronization. Our model is based on the architecture of one reference clock and many slave clocks communicating with each other over the Internet, and can be easily expanded to multi-tier tree architecture. The scalability for this kind of system is increased. The paper also discussed the standard issue of leader election in internal clock synchronization, and introduced a concept of clock priority, which consisted of Time Value Difference and Clock Precision Difference. A virtual ring was introduced in order to manage ring efficiently, and the policy of ring management was discussed. Especially when the leader node could not work, the system is how to find a new leader as soon as possible. In addition, we also present a node self-adaptive model, and induce a relationship between the clock precision and synchronization time; hence a node can predict when it should begin the synchronization process. When the communication link between the reference node and this slave node has crashed, salve node can continue to tune this time by clock precision difference itself and run normally.The method of clock direct adjustment by instantaneous time can cause a time discontinuity problem. This paper not only discusses backward correction and forward correction to adjust clock, but also the issues of clock adjustment. A concept of smallest scheduling unit was introduced for practical applications. When we use this interval to tune the clock step by step, the events or tasks will not lose the opportunity to be started or finished. An algorithm of clock self-adaptation and clock correction was described later in this chapter. Our chief objective in the thesis is to acquire better theoretical understanding of clock synchronization theory and method. A practical solution is discussed by simulation experimentations. Many of these tools, middleware and applications on grid require an identical clock system. In fact accurate global clock synchronization is invaluable for many applications –grid computing, distributed computing and applications.