Yong-Il Jang

RSP-DS: Real Time Sequential Pattern Analysis over Data Streams

The existing pattern analysis algorithms in data streams environment have only focused on studying performance improvement and effective memory usage. But when new data streams come, existing pattern analysis algorithms have to analyze patterns again and have to regenerate pattern tree. This approach needs many calculations in real time environments having real time pattern analysis needs. This paper proposes a method that continuously analyzes patterns of incoming data streams in real time. The proposed method analyzes patterns first, and then after obtains real time patterns by updating previously analyzed patterns. The patterns form a pattern tree, and freshly created new patterns update the pattern tree. In this way, real time patterns are always maintained in the pattern tree and old patterns in the tree are deleted easily using FIFO method. The advantage of our algorithm is proved by performance comparison with existing methods, MILE, with a condition that pattern is changed continuously.

A Load Balancing Method Using Ring Network in the Grid Database

In this paper, a load balancing method using ring network in the Grid database is proposed. Past research proposed to solve unbalanced load problem. But, past techniques can not be applied as the Grid database has a number of systems and user’s request always changes dynamically. The proposed method connects each node having the same replicated data through ring network. If workload overflows in some node, user’s request is transferred to a linked node which has the target data. And, this node stops another request from processing until the workload has significantly decreased. Then, to stop request forwarding from a previous node, it changes the link structure by sending a message to the previous node. Through performance evaluation, this paper shows that the proposed method has increased performance and is more suitable to the Grid database than the existing methods.

An extendible hashing based recovery method in a shared-nothing spatial database cluster

In this paper, a recovery method using extendible hashing in a shared-nothing spatial database cluster is proposed. The purpose is to increase the recovery performance and to decrease overhead of the system. In the case of failure, the recovery method in a database cluster restores the database using replicated data from neighbor node. When detect a failure, neighbor node writes the cluster log, and it must be transferred to a failure node. However, in neighbor node, one transaction makes several logs, and increase transferring log size. Also, this increases the recovery time and overhead of the internal network. The proposed method defines a novel cluster log that is composed of update type and a pointer to a record through RID or primary key. This is managed by extendible hashing in main memory. The last transaction replaces the cluster log. Through sending of last updated data, the number of cluster logs and transaction count in failure node are decreased. As a result, the method in this paper increased the availability of the database cluster.

CLUR-Tree for supporting frequent updates of data stream over sensor networks

Data streams from sensors are usually characterized as continuous, with very frequent updates. Queries over those data streams need to be processed in near real-time. So it is needed to design the index structure for supporting the frequent updates and fast retrieval of data efficiently. In this paper, CLUR-Tree (Cache-conscious Lazy Update R-Tree) is proposed, which is a spatial index for efficient processing of frequent updates of data streams in locality preserving monitoring applications. CLUR-Tree has two characteristics. First, it excludes index reconstruction overhead by permitting modification of only the index node of the sensor which moves out of the corresponding MBR (Minimum Bound Rectangle). Second, it reduces the key spaces by applying new compression method for MBR used as key in R-Tree and by considering cache to prevent bottleneck due to speed difference between main memory and CPU. The experimental results indicate that the proposed CLUR-Tree enhances update performance and gives a good retrieval performance simultaneously.