Statistics Reference

Overview

SQL Query Stress collects detailed performance statistics during load tests by leveraging SQL Server’s built-in statistics features (SET STATISTICS IO and SET STATISTICS TIME). This page documents all collected metrics, how they are gathered, and how to interpret them.

Collected Metrics

All metrics are stored in the QueryOutput class (defined in LoadEngine.cs:535-560) and collected per query execution.

Core Metrics

Time

TimeSpan

Description: Total elapsed time for the query execution as measured by the client using System.Diagnostics.Stopwatch.Unit: TimeSpan (milliseconds precision)Interpretation: Includes network latency, data retrieval time, and all client-side processing. This is wall-clock time from when the query starts until it completes.Source: Client-side stopwatch measurement

CpuTime

int

Description: Total CPU time consumed by SQL Server executing the query.Unit: MillisecondsInterpretation: Actual CPU processing time on the server. Does not include wait time for I/O, locks, or other resources. A query with high CpuTime relative to ElapsedTime is CPU-bound.Source: Parsed from SET STATISTICS TIME ON outputCollection: Enabled when CollectTimeStats is true

ElapsedTime

int

Description: Total elapsed time as reported by SQL Server.Unit: MillisecondsInterpretation: Server-side measurement of query duration. Includes CPU time plus all wait time (I/O, locks, parallelism waits, etc.). Does not include network transmission time.Source: Parsed from SET STATISTICS TIME ON outputCollection: Enabled when CollectTimeStats is true

LogicalReads

int

Description: Total number of logical reads (8KB pages) performed by the query across all tables.Unit: Pages (8KB each)Interpretation: Number of data pages read from the buffer cache (includes both cached and disk-read pages). Higher values indicate more data being scanned. Reducing logical reads through better indexing improves performance.Source: Summed from SET STATISTICS IO ON output for all tablesCollection: Enabled when CollectIoStats is true

Finished

boolean

Description: Indicates whether this was the last iteration for the thread.Interpretation: Used internally to track test progress and determine when threads complete.

ActiveThreads

int

Description: Number of threads still running at the time this result was recorded.Interpretation: Useful for understanding load characteristics over time. Shows how many concurrent connections were active.

Exception

Description: Exception object if the query execution failed.Interpretation: Contains detailed error information. Null if the query succeeded. Check for connection errors, timeout errors, or SQL syntax errors.

How Statistics Are Collected

1. Enabling Statistics Collection

When CollectIoStats or CollectTimeStats is enabled in the configuration, SQL Query Stress executes the following commands before running queries:

SET STATISTICS IO ON;
SET STATISTICS TIME ON;

Implementation reference: LoadEngine.cs:139-144

2. InfoMessage Event Handling

SQL Server sends statistics as informational messages through the SqlConnection.InfoMessage event. SQL Query Stress attaches an event handler (GetInfoMessages method at LoadEngine.cs:343-365) to capture these messages.

3. Regex Pattern Matching

Statistics are extracted from InfoMessage text using compiled regular expressions:

Logical Reads Pattern

@"(?:Table (\'\'w{1,}\'|'#\w{1,}\'|'##\w{1,}\'). Scan count \d{1,}, logical reads )(\d{1,})"

Matches: Messages like:

Table 'Customers'. Scan count 1, logical reads 42, physical reads 0, ...

Extracted Value: The number after “logical reads” (42 in the example) Implementation: LoadEngine.cs:280

CPU and Elapsed Time Pattern

@"(?:SQL Server Execution Times:|SQL Server parse and compile time:)(?:\s{1,}CPU time = )(\d{1,})(?: ms,\s{1,}elapsed time = )(\d{1,})"

Matches: Messages like:

SQL Server Execution Times:
   CPU time = 156 ms,  elapsed time = 2034 ms.

Extracted Values:

First captured group: CPU time in milliseconds (156)
Second captured group: Elapsed time in milliseconds (2034)

Implementation: LoadEngine.cs:284-287

4. Accumulation

When multiple tables are accessed, logical reads are summed across all tables. CPU and elapsed time are accumulated across parse/compile and execution phases. Reference: LoadEngine.cs:347-363

Understanding Statistics IO Output

When SET STATISTICS IO ON is enabled, SQL Server provides detailed I/O information:

Table 'Orders'. Scan count 1, logical reads 42, physical reads 0, 
read-ahead reads 0, lob logical reads 0, lob physical reads 0, 
lob read-ahead reads 0.

What SQL Query Stress Tracks

Metric	Tracked	Description
Logical Reads	✅ Yes	Pages read from buffer cache
Physical Reads	❌ No	Pages read from disk
Read-ahead Reads	❌ No	Pages read by read-ahead mechanism
Scan Count	❌ No	Number of seeks/scans performed
LOB Reads	❌ No	Large object read operations

SQL Query Stress focuses on logical reads as the primary I/O metric because it represents total data accessed regardless of cache state.

Understanding Statistics Time Output

When SET STATISTICS TIME ON is enabled, SQL Server provides timing information:

SQL Server parse and compile time: 
   CPU time = 0 ms, elapsed time = 5 ms.

SQL Server Execution Times:
   CPU time = 156 ms,  elapsed time = 2034 ms.

Metric Interpretation

Metric	Description	High Value Indicates
CPU Time	Pure computation time	CPU-bound operations, calculations, aggregations
Elapsed Time	Total server-side time	I/O waits, lock waits, parallelism waits
Time (Client)	Client-measured time	Network latency, result set transfer

Common Patterns

CPU-Bound Query:

CPU time = 1500 ms, elapsed time = 1550 ms

CPU ≈ Elapsed Time
Query spent most time in computation
Optimize with better algorithms, indexing, or reduce data volume

I/O-Bound Query:

CPU time = 50 ms, elapsed time = 3000 ms

Elapsed Time >> CPU Time
Query spent most time waiting for I/O
Optimize with better indexing, caching, or faster storage

Lock-Contention Query:

CPU time = 10 ms, elapsed time = 5000 ms

Very low CPU, very high elapsed time
Query spent most time waiting for locks
Review isolation levels, add WITH (NOLOCK) if appropriate, or optimize locking strategy

Aggregated Statistics

During a load test, the UI displays aggregated statistics across all query executions:

Calculated Metrics

Total Queries Executed:

NumThreads × NumIterations

Average Elapsed Time:

Sum(Time) / Total Queries

Average Logical Reads:

Sum(LogicalReads) / Total Queries

Queries Per Second:

Total Queries / Total Test Duration

Error Rate:

(Queries with E != null) / Total Queries × 100%

Configuration Impact on Statistics

ForceDataRetrieval Setting

When false (default):

Executes query with ExecuteNonQuery()
Results are generated but not transmitted to client
Lower client-side Time measurements
More representative of write operations

When true:

Reads all result sets completely using ExecuteReader()
Forces full data retrieval and network transmission
Higher client-side Time measurements
More representative of read operations with data consumption

Implementation: LoadEngine.cs:416-433

Statistics Collection Overhead

Enabling statistics collection adds minimal overhead:

CollectIoStats: ~1-5% overhead (parsing InfoMessages)
CollectTimeStats: ~1-5% overhead (parsing InfoMessages)

The overhead is negligible compared to query execution time and provides valuable diagnostic information.

Best Practices

For Performance Testing

Enable both statistics types to get complete picture:

"CollectIoStats": true,
"CollectTimeStats": true

Use ForceDataRetrieval when testing read workloads:
```
"ForceDataRetrieval": true
```
Monitor logical reads to validate index effectiveness
Compare CPU vs Elapsed Time to identify bottleneck type

For Accurate Measurements

Run warm-up iterations before collecting final statistics
Use sufficient iterations to get statistically significant averages
Monitor ActiveThreads to ensure consistent concurrency
Check for exceptions (E field) that might skew results

Configuration File Format - Enable statistics collection settings
CLI Command Reference - Run tests from command line
Interpreting Results - Interpret test results in the UI

Get Started

Guides

Configuration

Advanced

Reference

Overview

Collected Metrics

Core Metrics

How Statistics Are Collected

1. Enabling Statistics Collection

2. InfoMessage Event Handling

3. Regex Pattern Matching

Logical Reads Pattern

CPU and Elapsed Time Pattern

4. Accumulation

Understanding Statistics IO Output

What SQL Query Stress Tracks

Understanding Statistics Time Output

Metric Interpretation

Common Patterns

Aggregated Statistics

Calculated Metrics

Configuration Impact on Statistics

ForceDataRetrieval Setting

Statistics Collection Overhead

Best Practices

For Performance Testing

For Accurate Measurements

Build docs developers (and LLMs) love

Get Started

Guides

Configuration

Advanced

Reference

​Overview

​Collected Metrics

​Core Metrics

​How Statistics Are Collected

​1. Enabling Statistics Collection

​2. InfoMessage Event Handling

​3. Regex Pattern Matching

​Logical Reads Pattern

​CPU and Elapsed Time Pattern

​4. Accumulation

​Understanding Statistics IO Output

​What SQL Query Stress Tracks

​Understanding Statistics Time Output

​Metric Interpretation

​Common Patterns

​Aggregated Statistics

​Calculated Metrics

​Configuration Impact on Statistics

​ForceDataRetrieval Setting

​Statistics Collection Overhead

​Best Practices

​For Performance Testing

​For Accurate Measurements

​Related

Build docs developers (and LLMs) love

Overview

Collected Metrics

Core Metrics

How Statistics Are Collected

1. Enabling Statistics Collection

2. InfoMessage Event Handling

3. Regex Pattern Matching

Logical Reads Pattern

CPU and Elapsed Time Pattern

4. Accumulation

Understanding Statistics IO Output

What SQL Query Stress Tracks

Understanding Statistics Time Output

Metric Interpretation

Common Patterns

Aggregated Statistics

Calculated Metrics

Configuration Impact on Statistics

ForceDataRetrieval Setting

Statistics Collection Overhead

Best Practices

For Performance Testing

For Accurate Measurements

Related