Thread Pool Pattern

# Thread Creation Cost Analysis
 
## Scenario: 10,000 requests/second
 
### Time Spent Just Creating Threads:
- Linux:   10,000 × 20μs = 200ms per second (20% of CPU capacity)
- JVM:     10,000 × 100μs = 1000ms per second (100%—IMPOSSIBLE)
 
### Memory Overhead (Assuming 8MB stack reservation):
- 10,000 threads × 8MB = 80GB virtual address space
- Even with demand paging, each thread commits ~64KB minimum
- 10,000 × 64KB = 640MB committed memory JUST for stacks
 
### Context Switch Overhead:
- Creating thread: ~2 context switches (user→kernel→user)
- If each thread runs briefly before blocking: +2 more switches
- 10,000 × 4 switches × ~5μs = 200ms additional overhead
 
## Total Overhead: 400ms+ per second just for thread management

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45

// ❌ ANTI-PATTERN: Thread-per-request
public class NaiveServer {
    private ServerSocket serverSocket;
    
    public void start(int port) throws IOException {
        serverSocket = new ServerSocket(port);
        System.out.println("Server listening on port " + port);
        
        while (true) {
            // Accept incoming connection
            Socket clientSocket = serverSocket.accept();
            
            // ⚠️ PROBLEM: Creates a new thread for EVERY request
            Thread handler = new Thread(() -> {
                handleRequest(clientSocket);
            });
            handler.start();
            
            // Thread is created, runs, and then garbage collected
            // This happens thousands of times per second under load
        }
    }
    
    private void handleRequest(Socket socket) {
        try (
            BufferedReader in = new BufferedReader(
                new InputStreamReader(socket.getInputStream())
            );
            PrintWriter out = new PrintWriter(
                socket.getOutputStream(), true
            )
        ) {
            String request = in.readLine();
            
            // Simulate some processing work (10ms)
            Thread.sleep(10);
            
            out.println("HTTP/1.1 200 OK\r
\r
Hello World");
        } catch (Exception e) {
            e.printStackTrace();
        }
    }
}

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101

import java.util.concurrent.CountDownLatch;
 
/**
 * Benchmark to measure actual thread creation overhead
 * in your specific JVM and environment.
 */
public class ThreadCreationBenchmark {
    
    public static void main(String[] args) throws Exception {
        int[] threadCounts = {100, 1000, 5000, 10000};
        
        // Warmup
        System.out.println("Warming up...");
        benchmarkThreadCreation(1000);
        
        System.out.println("
Thread Creation Benchmark Results:");
        System.out.println("==================================");
        
        for (int count : threadCounts) {
            double avgMicros = benchmarkThreadCreation(count);
            double throughput = 1_000_000.0 / avgMicros;
            
            System.out.printf(
                "Threads: %5d | Avg Creation: %7.2f μs | Max Throughput: %,.0f threads/sec%n",
                count, avgMicros, throughput
            );
        }
        
        System.out.println("
Memory Overhead Analysis:");
        System.out.println("=========================");
        analyzeMemoryOverhead();
    }
    
    static double benchmarkThreadCreation(int threadCount) throws Exception {
        Thread[] threads = new Thread[threadCount];
        CountDownLatch startLatch = new CountDownLatch(1);
        CountDownLatch doneLatch = new CountDownLatch(threadCount);
        
        long startTime = System.nanoTime();
        
        for (int i = 0; i < threadCount; i++) {
            threads[i] = new Thread(() -> {
                try {
                    startLatch.await();  // Wait for signal
                    doneLatch.countDown();
                } catch (InterruptedException e) {
                    Thread.currentThread().interrupt();
                }
            });
            threads[i].start();
        }
        
        long creationTime = System.nanoTime() - startTime;
        
        // Let threads run and complete
        startLatch.countDown();
        doneLatch.await();
        
        // Wait for all threads to fully terminate
        for (Thread t : threads) {
            t.join();
        }
        
        return (creationTime / 1000.0) / threadCount;  // Avg in microseconds
    }
    
    static void analyzeMemoryOverhead() {
        Runtime runtime = Runtime.getRuntime();
        
        System.gc();
        long baseMemory = runtime.totalMemory() - runtime.freeMemory();
        
        Thread[] threads = new Thread[1000];
        CountDownLatch latch = new CountDownLatch(1);
        
        for (int i = 0; i < 1000; i++) {
            final int idx = i;
            threads[i] = new Thread(() -> {
                try {
                    latch.await();
                } catch (InterruptedException e) {}
            });
            threads[i].start();
        }
        
        System.gc();
        long withThreads = runtime.totalMemory() - runtime.freeMemory();
        long perThread = (withThreads - baseMemory) / 1000;
        
        System.out.printf("Approximate memory per thread: %,d bytes%n", perThread);
        System.out.printf("Estimated memory for 10,000 threads: %,d MB%n", 
            perThread * 10000 / (1024 * 1024));
        
        latch.countDown();
        for (Thread t : threads) {
            try { t.join(); } catch (InterruptedException e) {}
        }
    }
}