Distributed Real-time Database Buffer Management, And Load Balancing Algorithm Research

Distributed real-time database was born to satisfy rising new demands on databases in new application fields such as computer integrated manufacturing, industrial process control, air traffic control, intelligence traffic, stock transaction monitoring system from the beginning of the 1980s and 90s and respond to the development of computer science and transformation of database technology. Distributed real-time database has become the hotspot of current research due to its actual application demands.Data buffering plays a significant role in database system. Because memory access time is far less than disk access time, gaining direct access to data from memory database will save much time in processing transactions. Many distributed real-time databases mainly choose memory databases to store their full or partial working versions to meet the real-time need. However, in actuality, data in real-time databases grow exponentially. When excessive data occupy the database capacity, it cannot be stored into memory database. Therefore, the provision of effective buffer management is particularly important. This paper advances layer-by-layer buffer replacement algorithm based on the study of buffer management algorithm of real-time database. This algorithm uses tree structure and endows data page some real-time nature (such as deadline and time stamp) in order to select appropriate replacement page in the process of buffer replacement, thus improving buffer hit ratio and reducing deadline miss ratio in real-time transactions.Load balancing technology is the key research focus area in studying distributed database system. Distributed database is contrasted with centralized database. Different nodes in distributed database can concurrently or parallel process one task or more tasks. Nevertheless, the differences in the processing capacity of these nodes or the size of these assigned tasks easily cause load imbalance among nodes. Load balancing algorithm is designed to provide the solution to the problem. For the time being, there are many load balancing algorithms. This paper has studied related algorithms and distributed real-time database environment and puts forward load balancing algorithm based on network prediction and one based on data replication technology. On the one hand, the use of network distance prediction technology provides evidence to load allocation and data transfer of the whole real-time database system. Predicted network distance among nodes can be used to select best nodes for real-time transaction processing in the real-time network environment and light-weight nodes nearest heavy-light nodes, which will save much time for real-time transaction processing and improve the ratio of finishing transactions to meet the deadline. On the other hand, load balancing algorithm based on data replication technology uses data replication in database and hotspot data processing in peer-to-peer network respectively. It copies hotspot data in recent period to destination replication node nearest original replication node at the opportune time among light-weight nodes in real-time database system, which provides better load balancing environment for later load allocation. The algorithms prove to be effective after their contrast and experiment.