diff --git a/internal/metrics/collector.go b/internal/metrics/collector.go
index b5c2cf1..4f6f953 100644
--- a/internal/metrics/collector.go
+++ b/internal/metrics/collector.go
@@ -1,17 +1,12 @@
 package metrics
 
 import (
-	"encoding/json"
 	"fmt"
 	"log"
 	"net/http"
-	"os"
-	"path"
-	"proxy-go/internal/cache"
 	"proxy-go/internal/config"
-	"proxy-go/internal/constants"
 	"proxy-go/internal/models"
-	"proxy-go/internal/monitor"
+	"proxy-go/internal/utils"
 	"runtime"
 	"sort"
 	"sync"
@@ -19,219 +14,125 @@ import (
 	"time"
 )
 
+// Collector 指标收集器
 type Collector struct {
-	startTime      time.Time
-	activeRequests int64
-	totalRequests  int64
-	totalErrors    int64
-	totalBytes     atomic.Int64
-	latencySum     atomic.Int64
-	pathStats      sync.Map
-	refererStats   sync.Map
-	statusStats    [6]atomic.Int64
-	latencyBuckets [10]atomic.Int64
-	recentRequests struct {
-		sync.RWMutex
-		items  [1000]*models.RequestLog
-		cursor atomic.Int64
-	}
-	cache     *cache.Cache
-	monitor   *monitor.Monitor
-	statsPool sync.Pool
+	startTime       time.Time
+	activeRequests  int64
+	totalRequests   int64
+	totalErrors     int64
+	totalBytes      int64
+	latencySum      int64
+	pathStats       sync.Map
+	statusCodeStats sync.Map
+	recentRequests  *models.RequestQueue
+	config          *config.Config
 }
 
-var globalCollector *Collector
-
-func InitCollector(config *config.Config) error {
-	globalCollector = &Collector{
-		startTime:      time.Now(),
-		pathStats:      sync.Map{},
-		statusStats:    [6]atomic.Int64{},
-		latencyBuckets: [10]atomic.Int64{},
-	}
-
-	// 初始化 cache
-	globalCollector.cache = cache.NewCache(constants.CacheTTL)
-
-	// 初始化监控器
-	globalCollector.monitor = monitor.NewMonitor(globalCollector)
-
-	// 初始化对象池
-	globalCollector.statsPool = sync.Pool{
-		New: func() interface{} {
-			return make(map[string]interface{}, 20)
-		},
-	}
-
-	// 设置最后保存时间
-	lastSaveTime = time.Now()
+var (
+	instance *Collector
+	once     sync.Once
+)
 
+// InitCollector 初始化收集器
+func InitCollector(cfg *config.Config) error {
+	once.Do(func() {
+		instance = &Collector{
+			startTime:      time.Now(),
+			recentRequests: models.NewRequestQueue(1000),
+			config:         cfg,
+		}
+	})
 	return nil
 }
 
+// GetCollector 获取收集器实例
 func GetCollector() *Collector {
-	return globalCollector
+	return instance
 }
 
+// BeginRequest 开始请求
 func (c *Collector) BeginRequest() {
 	atomic.AddInt64(&c.activeRequests, 1)
 }
 
+// EndRequest 结束请求
 func (c *Collector) EndRequest() {
 	atomic.AddInt64(&c.activeRequests, -1)
 }
 
+// RecordRequest 记录请求
 func (c *Collector) RecordRequest(path string, status int, latency time.Duration, bytes int64, clientIP string, r *http.Request) {
-	// 更新总请求数
 	atomic.AddInt64(&c.totalRequests, 1)
+	atomic.AddInt64(&c.totalBytes, bytes)
+	atomic.AddInt64(&c.latencySum, int64(latency))
 
-	// 更新总字节数
-	c.totalBytes.Add(bytes)
-
-	// 更新状态码统计
-	if status >= 100 && status < 600 {
-		c.statusStats[status/100-1].Add(1)
-	}
-
-	// 更新错误数
 	if status >= 400 {
 		atomic.AddInt64(&c.totalErrors, 1)
 	}
 
-	// 更新延迟分布
-	bucket := int(latency.Milliseconds() / 100)
-	if bucket < 10 {
-		c.latencyBuckets[bucket].Add(1)
+	// 更新状态码统计
+	statusKey := fmt.Sprintf("%d", status)
+	if value, ok := c.statusCodeStats.Load(statusKey); ok {
+		atomic.AddInt64(value.(*int64), 1)
+	} else {
+		var count int64 = 1
+		c.statusCodeStats.Store(statusKey, &count)
 	}
 
 	// 更新路径统计
-	if stats, ok := c.pathStats.Load(path); ok {
-		pathStats := stats.(*models.PathStats)
-		pathStats.Requests.Add(1)
+	if pathStats, ok := c.pathStats.Load(path); ok {
+		stats := pathStats.(models.PathMetrics)
+		atomic.AddInt64(&stats.RequestCount, 1)
 		if status >= 400 {
-			pathStats.Errors.Add(1)
+			atomic.AddInt64(&stats.ErrorCount, 1)
 		}
-		pathStats.Bytes.Add(bytes)
-		pathStats.LatencySum.Add(int64(latency))
+		atomic.AddInt64(&stats.TotalLatency, int64(latency))
+		atomic.AddInt64(&stats.BytesTransferred, bytes)
 	} else {
-		newStats := &models.PathStats{}
-		newStats.Requests.Add(1)
-		if status >= 400 {
-			newStats.Errors.Add(1)
+		stats := models.PathMetrics{
+			Path:             path,
+			RequestCount:     1,
+			ErrorCount:       int64(map[bool]int{true: 1, false: 0}[status >= 400]),
+			TotalLatency:     int64(latency),
+			BytesTransferred: bytes,
 		}
-		newStats.Bytes.Add(bytes)
-		newStats.LatencySum.Add(int64(latency))
-		c.pathStats.Store(path, newStats)
+		c.pathStats.Store(path, stats)
 	}
 
-	// 更新引用来源统计
-	if referer := r.Header.Get("Referer"); referer != "" {
-		if stats, ok := c.refererStats.Load(referer); ok {
-			stats.(*models.PathStats).Requests.Add(1)
-		} else {
-			newStats := &models.PathStats{}
-			newStats.Requests.Add(1)
-			c.refererStats.Store(referer, newStats)
-		}
-	}
-
-	// 记录最近的请求
-	log := &models.RequestLog{
+	// 记录最近请求
+	c.recentRequests.Push(models.RequestLog{
 		Time:      time.Now(),
 		Path:      path,
 		Status:    status,
-		Latency:   latency,
+		Latency:   int64(latency),
 		BytesSent: bytes,
 		ClientIP:  clientIP,
-	}
-
-	c.recentRequests.Lock()
-	cursor := c.recentRequests.cursor.Add(1) % 1000
-	c.recentRequests.items[cursor] = log
-	c.recentRequests.Unlock()
-
-	c.latencySum.Add(int64(latency))
-
-	// 更新错误统计
-	if status >= 400 {
-		c.monitor.RecordError()
-	}
-	c.monitor.RecordRequest()
-
-	// 检查延迟
-	c.monitor.CheckLatency(latency, bytes)
+	})
 }
 
+// GetStats 获取统计数据
 func (c *Collector) GetStats() map[string]interface{} {
-	// 先查缓存
-	if stats, ok := c.cache.Get("stats"); ok {
-		if statsMap, ok := stats.(map[string]interface{}); ok {
-			return statsMap
-		}
+	var mem runtime.MemStats
+	runtime.ReadMemStats(&mem)
+
+	// 计算平均延迟
+	avgLatency := float64(0)
+	if c.totalRequests > 0 {
+		avgLatency = float64(c.latencySum) / float64(c.totalRequests)
 	}
 
-	stats := c.statsPool.Get().(map[string]interface{})
-	defer c.statsPool.Put(stats)
+	// 收集状态码统计
+	statusCodeStats := make(map[string]int64)
+	c.statusCodeStats.Range(func(key, value interface{}) bool {
+		statusCodeStats[key.(string)] = atomic.LoadInt64(value.(*int64))
+		return true
+	})
 
-	var m runtime.MemStats
-	runtime.ReadMemStats(&m)
-
-	uptime := time.Since(c.startTime)
-	currentRequests := atomic.LoadInt64(&c.totalRequests)
-	currentErrors := atomic.LoadInt64(&c.totalErrors)
-	currentBytes := c.totalBytes.Load()
-
-	// 计算错误率
-	var errorRate float64
-	if currentRequests > 0 {
-		errorRate = float64(currentErrors) / float64(currentRequests)
-	}
-
-	// 基础指标
-	stats["uptime"] = uptime.String()
-	stats["active_requests"] = atomic.LoadInt64(&c.activeRequests)
-	stats["total_requests"] = currentRequests
-	stats["total_errors"] = currentErrors
-	stats["error_rate"] = errorRate
-	stats["total_bytes"] = currentBytes
-	stats["bytes_per_second"] = float64(currentBytes) / Max(uptime.Seconds(), 1)
-	stats["requests_per_second"] = float64(currentRequests) / Max(uptime.Seconds(), 1)
-
-	// 系统指标
-	stats["num_goroutine"] = runtime.NumGoroutine()
-	stats["memory_usage"] = FormatBytes(m.Alloc)
-
-	// 延迟指标
-	latencySum := c.latencySum.Load()
-	if currentRequests > 0 {
-		stats["avg_response_time"] = FormatDuration(time.Duration(latencySum / currentRequests))
-	} else {
-		stats["avg_response_time"] = FormatDuration(0)
-	}
-
-	// 状态码统计
-	statusStats := make(map[string]int64)
-	for i := range c.statusStats {
-		statusStats[fmt.Sprintf("%dxx", i+1)] = c.statusStats[i].Load()
-	}
-	stats["status_code_stats"] = statusStats
-
-	// 延迟百分位数
-	stats["latency_percentiles"] = make([]float64, 0)
-
-	// 路径统计
+	// 收集路径统计
 	var pathMetrics []models.PathMetrics
 	c.pathStats.Range(func(key, value interface{}) bool {
-		stats := value.(*models.PathStats)
-		if stats.Requests.Load() > 0 {
-			pathMetrics = append(pathMetrics, models.PathMetrics{
-				Path:             key.(string),
-				RequestCount:     stats.Requests.Load(),
-				ErrorCount:       stats.Errors.Load(),
-				AvgLatency:       formatAvgLatency(stats.LatencySum.Load(), stats.Requests.Load()),
-				BytesTransferred: stats.Bytes.Load(),
-			})
-		}
+		stats := value.(models.PathMetrics)
+		pathMetrics = append(pathMetrics, stats)
 		return true
 	})
 
@@ -240,95 +141,32 @@ func (c *Collector) GetStats() map[string]interface{} {
 		return pathMetrics[i].RequestCount > pathMetrics[j].RequestCount
 	})
 
+	// 只保留前10个
 	if len(pathMetrics) > 10 {
-		stats["top_paths"] = pathMetrics[:10]
-	} else {
-		stats["top_paths"] = pathMetrics
+		pathMetrics = pathMetrics[:10]
 	}
 
-	// 最近请求
-	stats["recent_requests"] = c.getRecentRequests()
-
-	// 引用来源统计
-	var refererMetrics []models.PathMetrics
-	c.refererStats.Range(func(key, value interface{}) bool {
-		stats := value.(*models.PathStats)
-		if stats.Requests.Load() > 0 {
-			refererMetrics = append(refererMetrics, models.PathMetrics{
-				Path:         key.(string),
-				RequestCount: stats.Requests.Load(),
-			})
-		}
-		return true
-	})
-
-	// 按请求数排序
-	sort.Slice(refererMetrics, func(i, j int) bool {
-		return refererMetrics[i].RequestCount > refererMetrics[j].RequestCount
-	})
-
-	if len(refererMetrics) > 10 {
-		stats["top_referers"] = refererMetrics[:10]
-	} else {
-		stats["top_referers"] = refererMetrics
-	}
-
-	// 检查告警
-	c.monitor.CheckMetrics(stats)
-
-	// 写入缓存
-	c.cache.Set("stats", stats)
-
-	return stats
-}
-
-func (c *Collector) getRecentRequests() []models.RequestLog {
-	var recentReqs []models.RequestLog
-	c.recentRequests.RLock()
-	defer c.recentRequests.RUnlock()
-
-	cursor := c.recentRequests.cursor.Load()
-	for i := 0; i < 10; i++ {
-		idx := (cursor - int64(i) + 1000) % 1000
-		if c.recentRequests.items[idx] != nil {
-			recentReqs = append(recentReqs, *c.recentRequests.items[idx])
+	// 计算每个路径的平均延迟
+	for i := range pathMetrics {
+		if pathMetrics[i].RequestCount > 0 {
+			avgLatencyMs := float64(pathMetrics[i].TotalLatency) / float64(pathMetrics[i].RequestCount) / float64(time.Millisecond)
+			pathMetrics[i].AvgLatency = fmt.Sprintf("%.2fms", avgLatencyMs)
 		}
 	}
-	return recentReqs
-}
 
-// 辅助函数
-func FormatDuration(d time.Duration) string {
-	if d < time.Millisecond {
-		return fmt.Sprintf("%.2f μs", float64(d.Microseconds()))
+	return map[string]interface{}{
+		"uptime":            time.Since(c.startTime).String(),
+		"active_requests":   atomic.LoadInt64(&c.activeRequests),
+		"total_requests":    atomic.LoadInt64(&c.totalRequests),
+		"total_errors":      atomic.LoadInt64(&c.totalErrors),
+		"total_bytes":       atomic.LoadInt64(&c.totalBytes),
+		"num_goroutine":     runtime.NumGoroutine(),
+		"memory_usage":      utils.FormatBytes(int64(mem.Alloc)),
+		"avg_response_time": fmt.Sprintf("%.2fms", avgLatency/float64(time.Millisecond)),
+		"status_code_stats": statusCodeStats,
+		"top_paths":         pathMetrics,
+		"recent_requests":   c.recentRequests.GetAll(),
 	}
-	if d < time.Second {
-		return fmt.Sprintf("%.2f ms", float64(d.Milliseconds()))
-	}
-	return fmt.Sprintf("%.2f s", d.Seconds())
-}
-
-func FormatBytes(bytes uint64) string {
-	const (
-		MB = 1024 * 1024
-		KB = 1024
-	)
-
-	switch {
-	case bytes >= MB:
-		return fmt.Sprintf("%.2f MB", float64(bytes)/MB)
-	case bytes >= KB:
-		return fmt.Sprintf("%.2f KB", float64(bytes)/KB)
-	default:
-		return fmt.Sprintf("%d Bytes", bytes)
-	}
-}
-
-func Max(a, b float64) float64 {
-	if a > b {
-		return a
-	}
-	return b
 }
 
 func (c *Collector) SaveMetrics(stats map[string]interface{}) error {
@@ -354,7 +192,7 @@ func (c *Collector) validateLoadedData() error {
 	// 验证基础指标
 	if atomic.LoadInt64(&c.totalRequests) < 0 ||
 		atomic.LoadInt64(&c.totalErrors) < 0 ||
-		c.totalBytes.Load() < 0 {
+		atomic.LoadInt64(&c.totalBytes) < 0 {
 		return fmt.Errorf("invalid stats values")
 	}
 
@@ -364,20 +202,18 @@ func (c *Collector) validateLoadedData() error {
 	}
 
 	// 验证状态码统计
-	for i := range c.statusStats {
-		if c.statusStats[i].Load() < 0 {
-			return fmt.Errorf("invalid status code count at index %d", i)
-		}
-	}
+	c.statusCodeStats.Range(func(key, value interface{}) bool {
+		return atomic.LoadInt64(value.(*int64)) >= 0
+	})
 
 	// 验证路径统计
 	var totalPathRequests int64
 	c.pathStats.Range(func(_, value interface{}) bool {
-		stats := value.(*models.PathStats)
-		if stats.Requests.Load() < 0 || stats.Errors.Load() < 0 {
+		stats := value.(models.PathMetrics)
+		if stats.RequestCount < 0 || stats.ErrorCount < 0 {
 			return false
 		}
-		totalPathRequests += stats.Requests.Load()
+		totalPathRequests += stats.RequestCount
 		return true
 	})
 
@@ -389,69 +225,6 @@ func (c *Collector) validateLoadedData() error {
 	return nil
 }
 
-func formatAvgLatency(latencySum, requests int64) string {
-	if requests <= 0 || latencySum <= 0 {
-		return "0 ms"
-	}
-	return FormatDuration(time.Duration(latencySum / requests))
-}
-
-// SaveBackup 保存数据备份
-func (c *Collector) SaveBackup() error {
-	stats := c.GetStats()
-	backupFile := fmt.Sprintf("backup_%s.json", time.Now().Format("20060102_150405"))
-
-	data, err := json.MarshalIndent(stats, "", "  ")
-	if err != nil {
-		return err
-	}
-
-	return os.WriteFile(path.Join("data/backup", backupFile), data, 0644)
-}
-
-// LoadBackup 加载备份数据
-func (c *Collector) LoadBackup(backupFile string) error {
-	data, err := os.ReadFile(path.Join("data/backup", backupFile))
-	if err != nil {
-		return err
-	}
-
-	var stats map[string]interface{}
-	if err := json.Unmarshal(data, &stats); err != nil {
-		return err
-	}
-
-	return c.RestoreFromBackup(stats)
-}
-
-// RestoreFromBackup 从备份恢复数据
-func (c *Collector) RestoreFromBackup(stats map[string]interface{}) error {
-	// 恢复基础指标
-	if totalReqs, ok := stats["total_requests"].(int64); ok {
-		atomic.StoreInt64(&c.totalRequests, totalReqs)
-	}
-	if totalErrs, ok := stats["total_errors"].(int64); ok {
-		atomic.StoreInt64(&c.totalErrors, totalErrs)
-	}
-	if totalBytes, ok := stats["total_bytes"].(int64); ok {
-		c.totalBytes.Store(totalBytes)
-	}
-
-	// 恢复状态码统计
-	if statusStats, ok := stats["status_code_stats"].(map[string]int64); ok {
-		for group, count := range statusStats {
-			if len(group) > 0 {
-				idx := (int(group[0]) - '0') - 1
-				if idx >= 0 && idx < len(c.statusStats) {
-					c.statusStats[idx].Store(count)
-				}
-			}
-		}
-	}
-
-	return nil
-}
-
 // GetLastSaveTime 实现 interfaces.MetricsCollector 接口
 var lastSaveTime time.Time
 
diff --git a/internal/models/metrics.go b/internal/models/metrics.go
index 7a0a3eb..c58dffe 100644
--- a/internal/models/metrics.go
+++ b/internal/models/metrics.go
@@ -2,18 +2,8 @@ package models
 
 import (
 	"sync/atomic"
-	"time"
 )
 
-type RequestLog struct {
-	Time      time.Time
-	Path      string
-	Status    int
-	Latency   time.Duration
-	BytesSent int64
-	ClientIP  string
-}
-
 type PathStats struct {
 	Requests   atomic.Int64
 	Errors     atomic.Int64
@@ -29,11 +19,3 @@ type HistoricalMetrics struct {
 	ErrorRate     float64 `json:"error_rate"`
 	AvgLatency    float64 `json:"avg_latency"`
 }
-
-type PathMetrics struct {
-	Path             string `json:"path"`
-	RequestCount     int64  `json:"request_count"`
-	ErrorCount       int64  `json:"error_count"`
-	AvgLatency       string `json:"avg_latency"`
-	BytesTransferred int64  `json:"bytes_transferred"`
-}
diff --git a/internal/models/request.go b/internal/models/request.go
new file mode 100644
index 0000000..ccf95c2
--- /dev/null
+++ b/internal/models/request.go
@@ -0,0 +1,64 @@
+package models
+
+import (
+	"sync"
+	"time"
+)
+
+// RequestLog 请求日志
+type RequestLog struct {
+	Time      time.Time `json:"time"`
+	Path      string    `json:"path"`
+	Status    int       `json:"status"`
+	Latency   int64     `json:"latency"`
+	BytesSent int64     `json:"bytes_sent"`
+	ClientIP  string    `json:"client_ip"`
+}
+
+// PathMetrics 路径指标
+type PathMetrics struct {
+	Path             string `json:"path"`
+	RequestCount     int64  `json:"request_count"`
+	ErrorCount       int64  `json:"error_count"`
+	TotalLatency     int64  `json:"-"`
+	AvgLatency       string `json:"avg_latency"`
+	BytesTransferred int64  `json:"bytes_transferred"`
+}
+
+// RequestQueue 请求队列
+type RequestQueue struct {
+	sync.RWMutex
+	items  []RequestLog
+	size   int
+	cursor int
+}
+
+// NewRequestQueue 创建新的请求队列
+func NewRequestQueue(size int) *RequestQueue {
+	return &RequestQueue{
+		items: make([]RequestLog, size),
+		size:  size,
+	}
+}
+
+// Push 添加请求日志
+func (q *RequestQueue) Push(log RequestLog) {
+	q.Lock()
+	defer q.Unlock()
+	q.items[q.cursor] = log
+	q.cursor = (q.cursor + 1) % q.size
+}
+
+// GetAll 获取所有请求日志
+func (q *RequestQueue) GetAll() []RequestLog {
+	q.RLock()
+	defer q.RUnlock()
+	result := make([]RequestLog, 0, q.size)
+	for i := 0; i < q.size; i++ {
+		idx := (q.cursor - i - 1 + q.size) % q.size
+		if !q.items[idx].Time.IsZero() {
+			result = append(result, q.items[idx])
+		}
+	}
+	return result
+}