---
id: wiki-2026-0508-garbage-collection
title: Garbage Collection
category: 10_Wiki/Topics
status: verified
canonical_id: self
aliases: [GC, Garbage Collector, Memory Management]
duplicate_of: none
source_trust_level: A
confidence_score: 0.9
verification_status: applied
tags: [runtime, memory, performance, vm]
raw_sources: []
last_reinforced: 2026-05-10
github_commit: pending
tech_stack:
  language: java
  framework: jvm
---

# Garbage Collection

## 매 한 줄
> **"매 unreachable memory를 자동으로 reclaim한다"**. 매 GC는 manual malloc/free의 cognitive burden을 제거 — McCarthy의 Lisp(1959)에서 시작, modern V8/HotSpot/Go의 generational + concurrent collector로 evolution. 2026 trade-off는 매 throughput vs latency vs memory overhead.

## 매 핵심

### 매 Algorithm 분류
- **Mark-and-sweep**: 매 reachable mark → unreachable sweep (fragmentation)
- **Copying (semi-space)**: 매 live obj 새 space로 copy (50% memory waste)
- **Mark-compact**: 매 mark 후 live obj 한쪽 끝으로 compact
- **Reference counting**: 매 ref count 0 즉시 free (cycle 문제)

### 매 Generational Hypothesis
- **매 most objects die young**: 매 short-lived 95%+
- **Young gen** (eden + survivor): 매 frequent, fast minor GC
- **Old gen** (tenured): 매 infrequent, expensive major GC
- **매 write barrier**: old → young reference tracking

### 매 응용
1. JVM (G1, ZGC, Shenandoah) — sub-ms pause.
2. V8 (Orinoco) — concurrent + parallel + incremental.
3. Go (tricolor concurrent) — sub-ms STW pause.
4. .NET (Server GC, regions) — generational + LOH.

## 💻 패턴

### JVM ZGC 설정 (sub-ms pause)
```bash
java -XX:+UseZGC \
     -XX:+ZGenerational \
     -Xmx16g \
     -XX:SoftMaxHeapSize=14g \
     -jar app.jar
```

### Go GC tuning
```go
import "runtime/debug"

func init() {
    // 매 GOGC=100 default — 100% growth 시 trigger
    debug.SetGCPercent(50) // 매 lower = more frequent, less memory
    
    // 매 hard memory limit (Go 1.19+)
    debug.SetMemoryLimit(8 << 30) // 매 8GB
}

// 매 manual hint after large alloc
runtime.GC()
```

### V8 Heap snapshot (Node.js)
```javascript
const v8 = require('v8');
const fs = require('fs');

// 매 heap snapshot for leak analysis
const snapshot = v8.writeHeapSnapshot();
console.log(`Heap snapshot: ${snapshot}`);

// 매 heap statistics
console.log(v8.getHeapStatistics());
// { total_heap_size, used_heap_size, heap_size_limit, ... }
```

### Reference cycle break (Python)
```python
import weakref

class Parent:
    def __init__(self):
        self.children = []

class Child:
    def __init__(self, parent):
        # 매 strong cycle: parent ↔ child → leak
        # self.parent = parent
        # 매 fix: weakref
        self.parent = weakref.ref(parent)
    
    def get_parent(self):
        return self.parent()  # 매 None if collected
```

### Object pooling (avoid GC pressure)
```csharp
public class BulletPool
{
    private readonly Stack<Bullet> pool = new();
    
    public Bullet Rent()
    {
        if (pool.Count > 0) return pool.Pop();
        return new Bullet();
    }
    
    public void Return(Bullet b)
    {
        b.Reset();
        pool.Push(b);
    }
}
// 매 hot path에서 alloc 회피 → minor GC 감소
```

### Tricolor marking (concept, Go-style)
```go
// 매 white = unscanned, gray = in queue, black = scanned
// invariant: 매 black은 white를 직접 reference 안 함
// write barrier로 enforce:
func writeBarrier(slot **Object, ptr *Object) {
    if isBlack(slot) && isWhite(ptr) {
        markGray(ptr)  // 매 promote to gray
    }
    *slot = ptr
}
```

### .NET Span<T> (stack alloc, no GC)
```csharp
public int SumDigits(int n)
{
    Span<int> digits = stackalloc int[16];
    int i = 0;
    while (n > 0) { digits[i++] = n % 10; n /= 10; }
    int sum = 0;
    for (int j = 0; j < i; j++) sum += digits[j];
    return sum; // 매 zero heap allocation
}
```

## 매 결정 기준
| 상황 | GC choice |
|---|---|
| Low-latency trading (JVM) | ZGC / Shenandoah (sub-ms pause) |
| Throughput batch (JVM) | Parallel GC |
| Game engine (managed) | Object pooling + GC.Collect at safe points |
| Real-time embedded | Manual memory or Rust (no GC) |
| Server-side Go | Default tricolor + GOGC tuning |

**기본값**: 매 modern runtime의 default GC. 매 measure first (allocation profiler), tune later.

## 🔗 Graph
- 부모: [[Memory Management]]
- 변형: [[Reference Counting]]
- 응용: [[JVM]] · [[V8]]
- Adjacent: [[ARC]]

## 🤖 LLM 활용
**언제**: 매 GC pause 분석, allocation hotspot 식별, runtime flag tuning 추천.
**언제 X**: 매 Rust/C/C++ — GC 없음. 매 hard real-time — non-deterministic pause unacceptable.

## ❌ 안티패턴
- **Calling GC.Collect() 빈번**: 매 fragmentation + throughput 손실.
- **Finalizer 의존**: 매 non-deterministic, GC overhead 증가 → IDisposable 사용.
- **Large object heap thrash**: 매 LOH (.NET) 매번 alloc → fragmentation.
- **String concat in loop**: 매 immutable string × N alloc → StringBuilder 사용.

## 🧪 검증 / 중복
- Verified (Jones, Hosking, Moss, "The Garbage Collection Handbook", 2nd ed., 2023).
- Verified (Oracle ZGC docs, OpenJDK).
- 신뢰도 A.

## 🕓 Changelog
| 날짜 | 변경 |
|---|---|
| 2026-05-08 | Phase 1 |
| 2026-05-10 | Manual cleanup — generational GC + V8/Go/JVM modern collectors |