Brajendra Panda

Reconstructing the Database after Electronic Attacks

In spite of all existing security mechanisms, it is quite difficult to protect databases from electronic attacks. This research provides techniques to make an assessment of the damaged data and then to recover the affected data to consistent states after an attack is detected. Damage assessment is done using data dependency approach in order to obtain precise information on the damaged part of the database. Two algorithms are presented in this paper. The first algorithm performs the damage assessment and recovery simultaneously; whereas the second algorithm separates these two processes for improved efficiency. Both algorithms allow blind- writes on data items allowing damaged items to be recovered automatically.

Reorganization of the database log for information warfare data recovery

Using traditional logs that contain only before and after images of data items and record neither any of the read operations nor actual mathematical or logical operations associated with an update, a complete damage assessment and repair of the database is not possible in the post Information Warfare scenario. In this research, a new logging protocol is proposed which records all appropriate information required for the complete repair of a database that was updated by committed but affected transactions. This information includes various predicates and statements used in the transaction. An algorithm to incorporate these in the log is offered. Based on this new log, a method to perform damage assessment and recovery is presented.

Data dependency based logging for defensive information warfare

Recovery from information attacks is a difficult task as database management systems are not designed to deal with malicious committed transactions, Few existing methods developed for this purpose heavily rely on logs and require that the log must not be purged. This causes the log grow tremendously and, therefore, making recovery a prolonged process since scanning the log takes enormous amount o f time. Hence, it is necessary to identi~ and skip parts o f logs that contain unaffected operations. In this research, we have used data dependency approach to divide a log into multiple segments, each segment containing only related operations. We have presented the model and the algorithm for log clustering which will significantly enhance the performance of database recovery for defensive information warfare.

Transaction fusion in the wake of information warfare

In case of an information attack on databases it is vital to start the recovery effort as soon as an attack is identified. The existing recovery techniques first undo all malicious and affected transactions and then redo all affected transactions. This is too time-consuming. In this work, we have developed a model to fuse groups of malicious and affected transactions. During the undo process all new fused transactions are undone. Then, only fused affected transactions are re-executed to bring the database to a consistent state. This offers the following major benefits. First, by combining transactions, the number of operations such as start, commit, read, and write are reduced. As a result, several data items that were required to be accessed multiple times in each individual transaction are now accessed only once in a fused transaction. Moreover, the amount of log access is also reduced. This helps in accelerating the recovery process.

Design and analysis of data structures for querying image databases

Searching an image database for similar images has been a research issue for past several years. Many indexing structures such as R*-tree, (k, u) R-tree were proposed to improve search time. However, these indexing structures cannot efficiently handle the partial image-matching problem. Due to the absence of one or more expected objects, the searching process becomes extremely difficult. In this study, a new tree structure is developed for handling partial-matching problem efficiently. The proposed tree has only three levels, which are independent of the total number of images. Furthermore, the space used to store our index structure is very less. This helps in accelerating queries. We have also developed another tree structure, which shows excellent performance when all objects in the image are present. An analytical model is developed to verify the performance of these proposed methods.

Improving the performance of lineage tracing in data warehouse

The primary use of data warehouse is for decision support. Data warehouse mostly consists of derived data, generally stored as views, rather than raw data, to help in decision making. Sometimes decision-makers want to know the original tuples from which the view has been derived. The purpose of this may be to determine the authenticity of the original data or to bolster the decision-making capabilities. This problem of retrieving the original tuples of a view is termed as data lineage problem or lineage tracing problem. In our work we are suggesting a method, which will perform the task of data lineage with significantly less time. Our method suggests adding an auxiliary column to all tables in the database to store surrogate values for each tuple. Although this method requires extra disc space, the significant saving in response time outperforms this shortcoming.

An overview of post information warfare data recovery

This research presents a synopsis of a method to make an assessment of the damage made to the database in an information warfare situation. The protocol requires a modified log to store necessary information to re-compute the affected operations of benign transactions. After an attack is detected, the algorithm presented in this paper scans the log to assess the damage and then recovers the damaged data to a consistent state.