Performance Optimization and Testing

Load Runner allows manual script creation using its C-based scripting API without recording. You can write scripts from scratch using VuGen (Virtual User Generator) by coding the functions directly. This is useful when you can't record (no access to application) or need more control than recording provides. The scripting approach provides full flexibility to customize test scenarios programmatically.

Performance testing is a broad category of testing. Load testing, volume testing, and stress testing are all specific sub-types of performance testing designed to evaluate system behavior under various conditions.

If CPU utilization is u%, the available CPU capacity is (100-u)%. With CPU time t, the total response time including queuing/waiting is t divided by the utilization fraction: t / ((100-u)/100) = t*100/(100-u). This formula captures queueing theory where response time explodes as utilization approaches 100%. Options B, C, D incorrectly apply the percentage.

Quicksort uses divide-and-conquer, recursively partitioning around a pivot element. Each level processes all n elements, and there are O(log n) levels on average (balanced partitions), giving n log(n) complexity. Best and average cases are n log(n), worst case (already sorted with poor pivot choice) degrades to n squared. Linear n and exponential e^n are incorrect.

B*Tree indexes are the most common index type for equality and range queries. Bitmap indexes are efficient for low-cardinality columns. Function-based indexes store the results of function calls to improve performance of queries using those functions. 'Bmap' is not a standard index type.

In B*Trees, leaf nodes are connected via a doubly linked list to allow efficient range queries and sequential traversal in both directions. Singly linked lists only support unidirectional traversal, and generic pointers or structures do not define the specific list traversal pattern.

SQL execution follows these phases: Parse (syntax/semantic validation), Bind (variable binding), Execute (query execution), and Fetch (returning results to client). 'Retrieve' is not typically listed as a separate step in standard SQL execution phases.

Hard parse is more expensive than soft parse because hard parse involves optimization, query plan generation, and row source generation. Soft parse only checks the shared pool for an existing execution plan and skips these costly steps when a plan is found.

In OLTP (Online Transaction Processing) applications, individual SQL statements should execute very quickly - typically less than 10 milliseconds - to maintain high transaction throughput and responsiveness. Values of 50ms, 100ms, or 150ms would be considered too slow for OLTP workloads.

Using hints in SQL statements is generally not advisable because they hard-code optimization assumptions that may become outdated as data distribution changes. Hints should only be used as a last resort when the optimizer consistently makes poor choices and other tuning methods have been exhausted.

Blind searches (queries without proper WHERE clauses or index utilization) should NOT be allowed as they cause full table scans and severe performance issues. Database systems should require meaningful search criteria to prevent inefficient queries that impact overall system performance.

Explain Plan and Trace are the correct first steps for SQL tuning. Explain Plan shows the execution plan and access paths. Trace provides detailed execution statistics. Parse and Bind are execution phases, not tuning tools - they describe how SQL is processed, not how to analyze or improve it.

Bind variables (parameters) should always be used in SQL statements instead of concatenating values. This prevents SQL injection attacks, improves query performance through cursor sharing, and enhances security by separating code from data.

To join N tables in SQL without causing a Cartesian product, you need a minimum of N-1 join conditions. For 3 tables, N-1 equals 2 join conditions. Fewer conditions result in an unwanted cross join, while more conditions are unnecessary for a basic join structure.

Modern Oracle optimizers are cost-based and generally choose the most efficient join methods based on statistics. Explicitly controlling join types (using hints) is rarely necessary and can degrade performance if not done carefully. The optimizer should be trusted unless there's a proven performance issue.

Updating only changed columns is better practice than updating all columns. It reduces undo/redo log generation, minimizes lock contention, improves performance, and makes auditing easier. Updating all columns unnecessarily consumes resources and can trigger cascading effects.

Always fetching all columns is bad practice. It increases network traffic, memory usage, and I/O. Use SELECT with specific columns needed for the query. This is called 'projection' and is fundamental to SQL performance optimization.

Jensor is a Java profiling tool developed by TCS (Tata Consultancy Services). JProfiler and YourKit are commercial profilers from other companies. DataXplor is not a standard Java profiler. Option C is correct - Jensor is TCS's proprietary Java profiling solution.

HP LoadRunner (now Micro Focus LoadRunner) has been the most widely used performance testing tool in enterprise environments for decades. It supports multiple protocols and comprehensive load testing scenarios. Options A, C, and D are legitimate tools but LoadRunner has the largest market share and enterprise adoption.