SDD-1 Algorithm Implementation

Transcription

1 National Institute of Technology Karnataka, Surathkal Project Report on SDD-1 Algorithm Implementation Under the Guidance of: Mr. Dr. Anantha Narayana (Professor) Submitted by: Mr. Vasanth Raja Chittampally (10IT05F) Department of Information Technology, National Institute of Technology Karnataka, Surathkal

2 Introduction SDD-1 Algorithm popularly known as System of Distributed Databases is a distributed query optimization algorithm using semijoin approach. SDD-1 algorithm is derived from an earlier method called Hill-Climbing algorithm, which has the distinction of being the first distributed query processing algorithm. This algorithm does not use semijoins, nor does it assume the data replication and fragmentation. This algorithm is devised for the wide area point to point networks. The cost of transferring the result to the final site is ignored. The Hill-Climbing algorithm proceeds as follows. The input to the algorithm includes the query graph, location of relations, and relation statistics. Following completion of the initial local processing, an initial feasible solution is selected which is a global execution schedule that includes all intersite communication. It is obtained by computing the cost of all the execution strategies that transfer the all required relations to a single candidate result site, and tehn choosing the least costly strategy. The Hill-Climbing algorithm is in the class of greedy algorithms, which start with an initial feasible solution and iteratively improve it. The main problem is that strategies with higer initial cost, which could nevertheless produce better overall benefits, are ignored. Furthermore, the algorithm may get stuck at a local minimum cost solution and fail to reach the global minimum. The Hill-Climbing algorithm has been substantially improved in SDD-1 in a number of ways. The improved version makes extensive use of semijoins. The objective function is expressed in terms of total communication time (local time and response time are not considered). Finally, the algorithm uses statistics on the database profile, where a profile is associated with a relation. The improved version also selects an initial feasible solution that is iteratively refined. The cost of semijoin is that of transferring the semijoin attributes A, Cost(R SJA S) = TMSG +TTR* size(projecta(s)) While its benefit is the cost of transferring irrelevant tuples of R( which is avoided by semijoin): Benefit(R SJA S) = (1-SFSJ(S.A))*size(R)*TTR The output of the algorithm is global strategy for executing the query.

3 Implementation Details The solution was implemented in Java. The core of the algorithm is a single function which takes care of calculating the Cost, Benefit and the difference between Benefit-Cost. Among all the semijoins, it finds the beneficial semijoins. The beneficial semijoins mean for which Benefit is greater than the cost. The solution selects the join which is having the MAX(Benefit-Cost) and keeps in the Execution Strategy ArrayList and removes from the actual join ArrayList. Performs the table statistics Input: Here user is provided with an interactive GUI to specify, the relations and size in the first table, two relations and connection among them and another table with the selectivity factor, size. Once the user clicks on the submit button, internally it computes all the calculations and produces the optimal strategy and the assembly line.

4 Output: The output is displayed in a separate window with having a jtextarea. The output shows the each iteration joining candidates, best join selected for that iteration and every table before the iteration and after the operation is done.

5 References [1] Query Processing in a System or Distributed Databases (SDD-1) PHILIP A. BERNSTEIN and NATHAN GOODMAN EUGENE WONG CHRISTOPHER L. REEVE JAMES B. ROTHNIE, Jr. ACM Transactions on Database Systems, Vol. 6, No. 4, December 1961 [2] Principles of Distributed Database Systems by Tamer Ozsu, Patrick Valduriez and Ssridhar