Apex Benchmarking Guide

Apex Benchmarking Guide

Performance testing is essential for writing efficient Apex code. This guide covers reliable benchmarking techniques and real-world performance data.

Sources:

• James Simone - Benchmarking Matters
• Dan Appleman - Advanced Apex Programming
• Justus van den Berg - Heap & CPU Optimization

Why Benchmark?

"Premature optimization is the root of all evil" - but informed optimization is essential. Benchmarking answers:

1. Which approach is faster? (Loop styles, data structures)
2. Will this scale? (200 records vs 10,000)
3. Where are the bottlenecks? (CPU, heap, SOQL)

Dan Appleman's Benchmarking Technique

The gold standard for Apex performance testing:

The Pattern

// Run in Anonymous Apex for consistent environment
Long startTime = System.currentTimeMillis();

// Your code to benchmark
for (Integer i = 0; i < 10000; i++) {
    // Operation being tested
}

Long endTime = System.currentTimeMillis();
System.debug('Duration: ' + (endTime - startTime) + 'ms');

Key Principles

Example: Complete Benchmark

// Comprehensive benchmark template
public class BenchmarkRunner {

    public static void compareMethods() {
        Integer iterations = 10000;

        // Warm-up run (JIT compilation)
        warmUp();

        // Method A
        Long startA = System.currentTimeMillis();
        for (Integer i = 0; i < iterations; i++) {
            methodA();
        }
        Long durationA = System.currentTimeMillis() - startA;

        // Method B
        Long startB = System.currentTimeMillis();
        for (Integer i = 0; i < iterations; i++) {
            methodB();
        }
        Long durationB = System.currentTimeMillis() - startB;

        // Results
        System.debug('Method A: ' + durationA + 'ms');
        System.debug('Method B: ' + durationB + 'ms');
        System.debug('Difference: ' + Math.abs(durationA - durationB) + 'ms');
        System.debug('Winner: ' + (durationA < durationB ? 'Method A' : 'Method B'));
    }

    private static void warmUp() {
        for (Integer i = 0; i < 100; i++) {
            methodA();
            methodB();
        }
    }

    private static void methodA() { /* Implementation A */ }
    private static void methodB() { /* Implementation B */ }
}

Real-World Benchmark Results

String Concatenation vs String.join()

From Justus van den Berg's testing:

// ❌ SLOW: String concatenation in loop (O(n²) string copies)
String result = '';
for (Account acc : accounts) {
    result += acc.Name + '\n';  // Creates new string each time!
}

// ✅ FAST: String.join() (O(n) single allocation)
List<String> names = new List<String>();
for (Account acc : accounts) {
    names.add(acc.Name);
}
String result = String.join(names, '\n');

Loop Performance Comparison

From Beyond the Cloud benchmarking (10,000 iterations):

// 🏆 FASTEST: While loop
Iterator<Account> iter = accounts.iterator();
while (iter.hasNext()) {
    Account acc = iter.next();
    // process
}

// ✅ GOOD: Traditional for loop
for (Integer i = 0; i < accounts.size(); i++) {
    Account acc = accounts[i];
    // process
}

// ⚠️ CONVENIENT BUT SLOWER: Enhanced for-each
for (Account acc : accounts) {
    // process - OK for small collections
}

Map vs List Lookup

// ❌ SLOW: List lookup
List<Id> processedIds = new List<Id>();
for (Account acc : accounts) {
    if (!processedIds.contains(acc.Id)) {  // O(n) each time!
        processedIds.add(acc.Id);
    }
}

// ✅ FAST: Set lookup
Set<Id> processedIds = new Set<Id>();
for (Account acc : accounts) {
    if (!processedIds.contains(acc.Id)) {  // O(1) constant time
        processedIds.add(acc.Id);
    }
}

Governor Limit Ceilings

Official Limits

Practical Thresholds

Runtime Limit Checking

public void processWithSafety(List<Account> accounts) {
    Integer cpuWarning = 8000;
    Integer heapWarning = 4800000;

    for (Account acc : accounts) {
        // Check before each operation
        if (Limits.getCpuTime() > cpuWarning) {
            System.debug(LoggingLevel.WARN,
                'CPU approaching limit: ' + Limits.getCpuTime() + 'ms');
            // Consider switching to async or chunking
            break;
        }

        if (Limits.getHeapSize() > heapWarning) {
            System.debug(LoggingLevel.WARN,
                'Heap approaching limit: ' + Limits.getHeapSize() + ' bytes');
            break;
        }

        processAccount(acc);
    }
}

Benchmarking Anti-Patterns

❌ Don't Do These

Benchmarking Checklist

Before optimizing, verify:

• Ran benchmark multiple times (3-5 runs)
• Used 1000+ iterations for micro-benchmarks
• Tested with production-scale data (200+ records)
• Included warm-up phase
• Ran in Anonymous Apex (not test context)
• Compared both approaches fairly
• Considered readability trade-offs

When NOT to Optimize

Sometimes clarity beats performance:

// More readable, negligible performance difference for small collections
for (Account acc : accounts) {
    acc.Description = acc.Name + ' - Updated';
}

// vs micro-optimized but harder to read
Iterator<Account> iter = accounts.iterator();
while (iter.hasNext()) {
    Account acc = iter.next();
    acc.Description = acc.Name + ' - Updated';
}

Rule of thumb: Optimize when:

1. Processing 200+ records regularly
2. Approaching governor limits
3. User experience is affected
4. Benchmarks show measurable improvement

Related Resources

• templates/benchmarking-template.cls - Ready-to-use benchmark template
• templates/cpu-heap-optimization.cls - Optimization patterns
• Apex Log Analyzer - Visual performance analysis