[C#] 一致性哈希（Consistent Hashing）的实现

网上没看到C#的一致性哈希的实现，所以这里提供一份。

同样的代码也在：http://code.google.com/p/consistent-hash/

source code:

using System; using System.Collections.Generic; using System.Linq; using System.Runtime.InteropServices; using System.Security.Cryptography; using System.Text; namespace ConsistentHash { public class MurmurHash2 { public static UInt32 Hash(Byte[] data) { return Hash(data, 0xc58f1a7b); } const UInt32 m = 0x5bd1e995; const Int32 r = 24; [StructLayout(LayoutKind.Explicit)] struct BytetoUInt32Converter { [FieldOffset(0)] public Byte[] Bytes; [FieldOffset(0)] public UInt32[] UInts; } public static UInt32 Hash(Byte[] data, UInt32 seed) { Int32 length = data.Length; if (length == 0) return 0; UInt32 h = seed ^ (UInt32)length; Int32 currentIndex = 0; // array will be length of Bytes but contains Uints // therefore the currentIndex will jump with +1 while length will jump with +4 UInt32[] hackArray = new BytetoUInt32Converter { Bytes = data }.UInts; while (length >= 4) { UInt32 k = hackArray[currentIndex++]; k *= m; k ^= k >> r; k *= m; h *= m; h ^= k; length -= 4; } currentIndex *= 4; // fix the length switch (length) { case 3: h ^= (UInt16)(data[currentIndex++] | data[currentIndex++] << 8); h ^= (UInt32)data[currentIndex] << 16; h *= m; break; case 2: h ^= (UInt16)(data[currentIndex++] | data[currentIndex] << 8); h *= m; break; case 1: h ^= data[currentIndex]; h *= m; break; default: break; } // Do a few final mixes of the hash to ensure the last few // bytes are well-incorporated. h ^= h >> 13; h *= m; h ^= h >> 15; return h; } } class ConsistentHash<T> { SortedDictionary<int, T> circle = new SortedDictionary<int, T>(); int _replicate = 100; //default _replicate count int[] ayKeys = null; //cache the ordered keys for better performance //it's better you override the GetHashCode() of T. //we will use GetHashCode() to identify different node. public void Init(IEnumerable<T> nodes) { Init(nodes, _replicate); } public void Init(IEnumerable<T> nodes, int replicate) { _replicate = replicate; foreach (T node in nodes) { this.Add(node, false); } ayKeys = circle.Keys.ToArray(); } public void Add(T node) { Add(node, true); } private void Add(T node, bool updateKeyArray) { for (int i = 0; i < _replicate; i++) { int hash = BetterHash(node.GetHashCode().ToString() + i); circle[hash] = node; } if (updateKeyArray) { ayKeys = circle.Keys.ToArray(); } } public void Remove(T node) { for (int i = 0; i < _replicate; i++) { int hash = BetterHash(node.GetHashCode().ToString() + i); if (!circle.Remove(hash)) { throw new Exception("can not remove a node that not added"); } } ayKeys = circle.Keys.ToArray(); } //we keep this function just for performance compare private T GetNode_slow(String key) { int hash = BetterHash(key); if (circle.ContainsKey(hash)) { return circle[hash]; } int first = circle.Keys.FirstOrDefault(h => h >= hash); if (first == new int()) { first = ayKeys[0]; } T node = circle[first]; return node; } //return the index of first item that >= val. //if not exist, return 0; //ay should be ordered array. int First_ge(int[] ay, int val) { int begin = 0; int end = ay.Length - 1; if (ay[end] < val || ay[0] > val) { return 0; } int mid = begin; while (end - begin > 1) { mid = (end + begin) / 2; if (ay[mid] >= val) { end = mid; } else { begin = mid; } } if (ay[begin] > val || ay[end] < val) { throw new Exception("should not happen"); } return end; } public T GetNode(String key) { //return GetNode_slow(key); int hash = BetterHash(key); int first = First_ge(ayKeys, hash); //int diff = circle.Keys[first] - hash; return circle[ayKeys[first]]; } //default String.GetHashCode() can't well spread strings like "1", "2", "3" public static int BetterHash(String key) { uint hash = MurmurHash2.Hash(Encoding.ASCII.GetBytes(key)); return (int)hash; } } }

example:

class Server { public int ID { get; set; } public Server(int id) { ID = id; } public override int GetHashCode() { return ID.GetHashCode(); } } private void btnTest_Click(object sender, EventArgs e) { List<Server> servers = new List<Server>(); for (int i = 0; i < 1000; i++) { servers.Add(new Server(i)); } ConsistentHash<Server> ch = new ConsistentHash<Server>(); ch.Init(servers); int search = 100000; DateTime start = DateTime.Now; SortedList<int, int> ay1 = new SortedList<int, int>(); for (int i = 0; i < search; i++) { int temp = ch.GetNode(i.ToString()).ID; ay1[i] = temp; } TimeSpan ts = DateTime.Now - start; MessageBox.Show(search + " each use macro seconds: " + (ts.TotalMilliseconds/search)*1000); //ch.Add(new Server(1000)); ch.Remove(servers[1]); SortedList<int, int> ay2 = new SortedList<int, int>(); for (int i = 0; i < search; i++) { int temp = ch.GetNode(i.ToString()).ID; ay2[i] = temp; } int diff = 0; for (int i = 0; i < search; i++) { if (ay1[i] != ay2[i]) { diff++; } } MessageBox.Show("diff: " + diff); }

性能测试结果：

不管是100个还是1000个server的情况，每次GetNode()耗时都是1到2微秒左右。应该是足够快了。