In this article we will discuss about the index concurrency control method for controlling database concurrency. Index as we know is a data structure that is used for easy navigation through the user data in a database. Index data should not be confused with user data. The difference between the two is that the former primarily consists of pointers.
Indexes have to be updated if any changes including delete, insert or modify are made to the database files so that user data can be accurately accessed. The index integrity is maintained by the means of a technique called the index locking. While a database transaction is taking place, lock is placed on a certain portion of the index. This is the portion that the transaction accesses in turn to access user data related to it. On top of this, for modifying and maintaining an index, special database system transactions are called. This is done by the system as a part of its self – maintenance routine. When a transaction locks a part of the index, the access to this portion is blocked to the other transactions. Thus, the other transactions cannot read or modify that portion. Only read operations can be performed if the lock is a shared one.
Indexes can be accessed using the techniques specializing in concurrency control. These techniques perform based upon the structure and type of the index. These techniques when applied to indexes are more effective when compared to application on user data. For B – trees we have specialized techniques that are effective in B – Tree concurrency control. For maintaining coordination between the threads that want to access the same indexes, index locks are used. The duration of the index locks is less than the duration of the usual transaction locks. Sometimes these index locks are also known as latches. Real time database systems are the ones that are most dependent on indexes for speeding up the access to data.
Index concurrency control also helps these systems in completing as many transactions as possible before deadline. For the prevention of the index contention so that it does not become a problem, we have special protocols called the high performance ICC (index concurrency control) protocols. By means of a detailed simulation model, real time variants of ICC protocols for B – Tree can be created and also their performances can be compared. GUARD – link is an ICC protocol for real time systems which can be both evaluated as well as presented. The classical B – link protocol is augmented with the admission control mechanism based upon feedback using this protocol. The ICC protocols are evaluated using certain performance metrics which are the percentage of the missed transactions. Sometimes the metric might be the percentage of the fairness w.r.t. type and size of the transaction.
According to some experiements, there is a difference between the real time ICC protocols’ performance characteristics and the performance characteristics of these same protocols in general purpose database systems. A thing to be noted about B link protocols is that they perform best when implemented in the conventional database systems, whereas in real time systems their performance is very poor since the load is too heavy. This GUARD – link protocol provides an all – new approach even though it has been developed on the grounds of B – link approach. It has been found after an experiment that this is the protocol that gives best performance under all conditions be it less or heavy real time workload. This is all because of its admission control mechanism.
