一致性 hash 算法（ consistent hashing ）

 

一致性hash算法 - consistent hashing

分类：  算法艺术2010-02-02 09:19  44257人阅读  评论(76)  收藏  举报

算法 cache object 服务器 存储 c

目录(?)[+]

一致性 hash 算法（ consistent hashing ）

张亮

consistent hashing 算法早在 1997 年就在论文 Consistent hashing and random trees 中被提出，目前在 cache 系统中应用越来越广泛；

1 基本场景

比如你有 N 个 cache 服务器（后面简称 cache ），那么如何将一个对象 object 映射到 N 个 cache 上呢，你很可能会采用类似下面的通用方法计算 object 的 hash 值，然后均匀的映射到到 N 个 cache ；

hash(object)%N

一切都运行正常，再考虑如下的两种情况；

1 一个 cache 服务器 m down 掉了（在实际应用中必须要考虑这种情况），这样所有映射到 cache m 的对象都会失效，怎么办，需要把 cache m 从 cache 中移除，这时候 cache 是 N-1 台，映射公式变成了 hash(object)%(N-1) ；

2 由于访问加重，需要添加 cache ，这时候 cache 是 N+1 台，映射公式变成了 hash(object)%(N+1) ；

1 和 2 意味着什么？这意味着突然之间几乎所有的 cache 都失效了。对于服务器而言，这是一场灾难，洪水般的访问都会直接冲向后台服务器；

再来考虑第三个问题，由于硬件能力越来越强，你可能想让后面添加的节点多做点活，显然上面的 hash 算法也做不到。

  有什么方法可以改变这个状况呢，这就是 consistent hashing...

2 hash 算法和单调性

　　 Hash 算法的一个衡量指标是单调性（ Monotonicity ），定义如下：

　　单调性是指如果已经有一些内容通过哈希分派到了相应的缓冲中，又有新的缓冲加入到系统中。哈希的结果应能够保证原有已分配的内容可以被映射到新的缓冲中去，而不会被映射到旧的缓冲集合中的其他缓冲区。

容易看到，上面的简单 hash 算法 hash(object)%N 难以满足单调性要求。

3 consistent hashing 算法的原理

consistent hashing 是一种 hash 算法，简单的说，在移除 / 添加一个 cache 时，它能够尽可能小的改变已存在 key 映射关系，尽可能的满足单调性的要求。

下面就来按照 5 个步骤简单讲讲 consistent hashing 算法的基本原理。

3.1 环形hash 空间

考虑通常的 hash 算法都是将 value 映射到一个 32 为的 key 值，也即是 0~2^32-1 次方的数值空间；我们可以将这个空间想象成一个首（ 0 ）尾（ 2^32-1 ）相接的圆环，如下面图 1 所示的那样。

图 1 环形 hash 空间

3.2 把对象映射到hash 空间

接下来考虑 4 个对象 object1~object4 ，通过 hash 函数计算出的 hash 值 key 在环上的分布如图 2 所示。

hash(object1) = key1;

… …

hash(object4) = key4;

图 2 4 个对象的 key 值分布

3.3 把cache 映射到hash 空间

Consistent hashing 的基本思想就是将对象和 cache 都映射到同一个 hash 数值空间中，并且使用相同的hash 算法。

假设当前有 A,B 和 C 共 3 台 cache ，那么其映射结果将如图 3 所示，他们在 hash 空间中，以对应的 hash值排列。

hash(cache A) = key A;

… …

hash(cache C) = key C;

图 3 cache 和对象的 key 值分布

 

说到这里，顺便提一下 cache 的 hash 计算，一般的方法可以使用 cache 机器的 IP 地址或者机器名作为hash 输入。

3.4 把对象映射到cache

现在 cache 和对象都已经通过同一个 hash 算法映射到 hash 数值空间中了，接下来要考虑的就是如何将对象映射到 cache 上面了。

在这个环形空间中，如果沿着顺时针方向从对象的 key 值出发，直到遇见一个 cache ，那么就将该对象存储在这个 cache 上，因为对象和 cache 的 hash 值是固定的，因此这个 cache 必然是唯一和确定的。这样不就找到了对象和 cache 的映射方法了吗？！

依然继续上面的例子（参见图 3 ），那么根据上面的方法，对象 object1 将被存储到 cache A 上； object2和 object3 对应到 cache C ； object4 对应到 cache B ；

3.5 考察cache 的变动

前面讲过，通过 hash 然后求余的方法带来的最大问题就在于不能满足单调性，当 cache 有所变动时，cache 会失效，进而对后台服务器造成巨大的冲击，现在就来分析分析 consistent hashing 算法。

3.5.1 移除 cache

考虑假设 cache B 挂掉了，根据上面讲到的映射方法，这时受影响的将仅是那些沿 cache B 逆时针遍历直到下一个 cache （ cache C ）之间的对象，也即是本来映射到 cache B 上的那些对象。

因此这里仅需要变动对象 object4 ，将其重新映射到 cache C 上即可；参见图 4 。

图 4 Cache B 被移除后的 cache 映射

3.5.2 添加 cache

再考虑添加一台新的 cache D 的情况，假设在这个环形 hash 空间中， cache D 被映射在对象 object2 和object3 之间。这时受影响的将仅是那些沿 cache D 逆时针遍历直到下一个 cache （ cache B ）之间的对象（它们是也本来映射到 cache C 上对象的一部分），将这些对象重新映射到 cache D 上即可。

 

因此这里仅需要变动对象 object2 ，将其重新映射到 cache D 上；参见图 5 。

图 5 添加 cache D 后的映射关系

4 虚拟节点

考量 Hash 算法的另一个指标是平衡性 (Balance) ，定义如下：

平衡性

　　平衡性是指哈希的结果能够尽可能分布到所有的缓冲中去，这样可以使得所有的缓冲空间都得到利用。

hash 算法并不是保证绝对的平衡，如果 cache 较少的话，对象并不能被均匀的映射到 cache 上，比如在上面的例子中，仅部署 cache A 和 cache C 的情况下，在 4 个对象中， cache A 仅存储了 object1 ，而 cache C 则存储了 object2 、 object3 和 object4 ；分布是很不均衡的。

为了解决这种情况， consistent hashing 引入了“虚拟节点”的概念，它可以如下定义：

“虚拟节点”（ virtual node ）是实际节点在 hash 空间的复制品（ replica ），一实际个节点对应了若干个“虚拟节点”，这个对应个数也成为“复制个数”，“虚拟节点”在 hash 空间中以 hash 值排列。

仍以仅部署 cache A 和 cache C 的情况为例，在图 4 中我们已经看到， cache 分布并不均匀。现在我们引入虚拟节点，并设置“复制个数”为 2 ，这就意味着一共会存在 4 个“虚拟节点”， cache A1, cache A2 代表了 cache A ； cache C1, cache C2 代表了 cache C ；假设一种比较理想的情况，参见图 6 。

图 6 引入“虚拟节点”后的映射关系

 

此时，对象到“虚拟节点”的映射关系为：

objec1->cache A2 ； objec2->cache A1 ； objec3->cache C1 ； objec4->cache C2 ；

因此对象 object1 和 object2 都被映射到了 cache A 上，而 object3 和 object4 映射到了 cache C 上；平衡性有了很大提高。

引入“虚拟节点”后，映射关系就从 { 对象 -> 节点 } 转换到了 { 对象 -> 虚拟节点 } 。查询物体所在 cache时的映射关系如图 7 所示。

图 7 查询对象所在 cache

 

“虚拟节点”的 hash 计算可以采用对应节点的 IP 地址加数字后缀的方式。例如假设 cache A 的 IP 地址为202.168.14.241 。

引入“虚拟节点”前，计算 cache A 的 hash 值：

Hash(“202.168.14.241”);

引入“虚拟节点”后，计算“虚拟节”点 cache A1 和 cache A2 的 hash 值：

Hash(“202.168.14.241#1”);  // cache A1

Hash(“202.168.14.241#2”);  // cache A2

5 小结

Consistent hashing 的基本原理就是这些，具体的分布性等理论分析应该是很复杂的，不过一般也用不到。

http://weblogs.java.net/blog/2007/11/27/consistent-hashing 上面有一个 java 版本的例子，可以参考。

http://blog.csdn.net/mayongzhan/archive/2009/06/25/4298834.aspx 转载了一个 PHP 版的实现代码。

http://www.codeproject.com/KB/recipes/lib-conhash.aspx C语言版本

 

一些参考资料地址：

http://portal.acm.org/citation.cfm?id=258660

http://en.wikipedia.org/wiki/Consistent_hashing

http://www.spiteful.com/2008/03/17/programmers-toolbox-part-3-consistent-hashing/

 http://weblogs.java.net/blog/2007/11/27/consistent-hashing

http://tech.idv2.com/2008/07/24/memcached-004/

http://blog.csdn.net/mayongzhan/archive/2009/06/25/4298834.aspx

 

 

Consistent hashing

By  sparkliang,  2 Feb 2010

4.74 (14 votes)

• Download source code - 20.3 KB

What is libconhash

libconhash is a consistent hashing library which can be compiled both on Windows and Linux platforms, with the following features:

1. High performance and easy to use, libconhash uses a red-black tree to manage all nodes to achieve high performance.
2. By default, it uses the MD5 algorithm, but it also supports user-defined hash functions.
3. Easy to scale according to the node's processing capacity.

Consistent hashing

Why you need consistent hashing

Now we will consider the common way to do load balance. The machine number chosen to cache object o will be:

 Collapse  |  Copy Code

hash(o) mod n

Here, n is the total number of cache machines. While this works well until you add or remove cache machines:

1. When you add a cache machine, then object o will be cached into the machine:
    Collapse  |  Copy Code

   hash(o) mod (n+1)

2. When you remove a cache machine, then object o will be cached into the machine:
    Collapse  |  Copy Code

   hash(o) mod (n-1)

So you can see that almost all objects will hashed into a new location. This will be a disaster since the originating content servers are swamped with requests from the cache machines. And this is why you need consistent hashing.

Consistent hashing can guarantee that when a cache machine is removed, only the objects cached in it will be rehashed; when a new cache machine is added, only a fairly few objects will be rehashed.

Now we will go into consistent hashing step by step.

Hash space

Commonly, a hash function will map a value into a 32-bit key, 0~2^32-1. Now imagine mapping the range into a circle, then the key will be wrapped, and 0 will be followed by 2^³²-1, as illustrated in figure 1.

Figure 1

Map object into hash space

Now consider four objects: object1~object4. We use a hash function to get their key values and map them into the circle, as illustrated in figure 2.

Figure 2

 Collapse  |  Copy Code

hash(object1) = key1;
.....
hash(object4) = key4;

Map the cache into hash space

The basic idea of consistent hashing is to map the cache and objects into the same hash space using the same hash function.

Now consider we have three caches, A, B and C, and then the mapping result will look like in figure 3.

 Collapse  |  Copy Code

hash(cache A) = key A;
....
hash(cache C) = key C;

Figure 3

Map objects into cache

Now all the caches and objects are hashed into the same space, so we can determine how to map objects into caches. Take object obj for example, just start from where obj is and head clockwise on the ring until you find a server. If that server is down, you go to the next one, and so forth. See figure 3 above.

According to the method, object1 will be cached into cache A; object2 and object3 will be cached into cache C, and object4 will be cached into cache B.

Add or remove cache

Now consider the two scenarios, a cache is down and removed; and a new cache is added.

If cache B is removed, then only the objects that cached in B will be rehashed and moved to C; in the example, seeobject4 illustrated in figure 4.

Figure 4

If a new cache D is added, and D is hashed between object2 and object3 in the ring, then only the objects that are between D and B will be rehashed; in the example, see object2, illustrated in figure 5.

Figure 5

Virtual nodes

It is possible to have a very non-uniform distribution of objects between caches if you don't deploy enough caches. The solution is to introduce the idea of "virtual nodes".

Virtual nodes are replicas of cache points in the circle, each real cache corresponds to several virtual nodes in the circle; whenever we add a cache, actually, we create a number of virtual nodes in the circle for it; and when a cache is removed, we remove all its virtual nodes from the circle.

Consider the above example. There are two caches A and C in the system, and now we introduce virtual nodes, and the replica is 2, then three will be 4 virtual nodes. Cache A1 and cache A2 represent cache A; cache C1 and cache C2 represent cache C, illustrated as in figure 6.

Figure 6

Then, the map from object to the virtual node will be:

 Collapse  |  Copy Code

objec1->cache A2; objec2->cache A1; objec3->cache C1; objec4->cache C2

When you get the virtual node, you get the cache, as in the above figure.

So object1 and object2 are cached into cache A, and object3 and object4 are cached into cache. The result is more balanced now.

So now you know what consistent hashing is.

Using the code

Interfaces of libconhash

 Collapse  |  Copy Code

/* initialize conhash library
 * @pfhash : hash function, NULL to use default MD5 method
 * return a conhash_s instance
 */
CONHASH_API struct conhash_s* conhash_init(conhash_cb_hashfunc pfhash);

/* finalize lib */
CONHASH_API void conhash_fini(struct conhash_s *conhash);

/* set node */
CONHASH_API void conhash_set_node(struct node_s *node, 
                 const char *iden, u_int replica);

/* 
 * add a new node 
 * @node: the node to add
 */
CONHASH_API int conhash_add_node(struct conhash_s *conhash, 
                                 struct node_s *node);

/* remove a node */
CONHASH_API int conhash_del_node(struct conhash_s *conhash,
                                 struct node_s *node);
...

/* 
 * lookup a server which object belongs to 
 * @object: the input string which indicates an object
 * return the server_s structure, do not modify the value, 
 * or it will cause a disaster
 */
CONHASH_API const struct node_s* 
  conhash_lookup(const struct conhash_s *conhash, 
  const char *object);

Libconhash is very easy to use. There is a sample in the project that shows how to use the library.

First, create a conhash instance. And then you can add or remove nodes of the instance, and look up objects.

The update node's replica function is not implemented yet.

 Collapse  |  Copy Code

/* init conhash instance */
struct conhash_s *conhash = conhash_init(NULL);
if(conhash)
{
    /* set nodes */
    conhash_set_node(&g_nodes[0], "titanic", 32);
    /* ... */

    /* add nodes */
    conhash_add_node(conhash, &g_nodes[0]);
    /* ... */
    printf("virtual nodes number %d\n", conhash_get_vnodes_num(conhash));
    printf("the hashing results--------------------------------------:\n");

    /* lookup object */
    node = conhash_lookup(conhash, "James.km");
    if(node) printf("[%16s] is in node: [%16s]\n", str, node->iden);
}

Reference

• http://portal.acm.org/citation.cfm?id=258660
• http://en.wikipedia.org/wiki/Consistent_hashing
• http://www.spiteful.com/2008/03/17/programmers-toolbox-part-3-consistent-hashing/

License

This article, along with any associated source code and files, is licensed under The BSD License

Consistent Hashing

Posted by  tomwhite on November 27, 2007 at 9:56 AM PST

I've bumped into consistent hashing a couple of times lately. The paper that introduced the idea (Consistent Hashing and Random Trees: Distributed Caching Protocols for Relieving Hot Spots on the World Wide Web by David Karger et al) appeared ten years ago, although recently it seems the idea has quietly been finding its way into more and more services, from Amazon's Dynamo to memcached (courtesy of Last.fm). So what is consistent hashing and why should you care?

The need for consistent hashing arose from limitations experienced while running collections of caching machines - web caches, for example. If you have a collection of n cache machines then a common way of load balancing across them is to put object o in cache machine number hash(o) mod n. This works well until you add or remove cache machines (for whatever reason), for then n changes andevery object is hashed to a new location. This can be catastrophic since the originating content servers are swamped with requests from the cache machines. It's as if the cache suddenly disappeared. Which it has, in a sense. (This is why you should care - consistent hashing is needed to avoid swamping your servers!)

It would be nice if, when a cache machine was added, it took its fair share of objects from all the other cache machines. Equally, when a cache machine was removed, it would be nice if its objects were shared between the remaining machines. This is exactly what consistent hashing does - consistently maps objects to the same cache machine, as far as is possible, at least.

The basic idea behind the consistent hashing algorithm is to hash both objects and caches using the same hash function. The reason to do this is to map the cache to an interval, which will contain a number of object hashes. If the cache is removed then its interval is taken over by a cache with an adjacent interval. All the other caches remain unchanged.

Demonstration

Let's look at this in more detail. The hash function actually maps objects and caches to a number range. This should be familiar to every Java programmer - the hashCode method on Object returns an int, which lies in the range -2³¹ to 2³¹-1. Imagine mapping this range into a circle so the values wrap around. Here's a picture of the circle with a number of objects (1, 2, 3, 4) and caches (A, B, C) marked at the points that they hash to (based on a diagram from Web Caching with Consistent Hashing by David Karger et al):

To find which cache an object goes in, we move clockwise round the circle until we find a cache point. So in the diagram above, we see object 1 and 4 belong in cache A, object 2 belongs in cache B and object 3 belongs in cache C. Consider what happens if cache C is removed: object 3 now belongs in cache A, and all the other object mappings are unchanged. If then another cache D is added in the position marked it will take objects 3 and 4, leaving only object 1 belonging to A.

This works well, except the size of the intervals assigned to each cache is pretty hit and miss. Since it is essentially random it is possible to have a very non-uniform distribution of objects between caches. The solution to this problem is to introduce the idea of "virtual nodes", which are replicas of cache points in the circle. So whenever we add a cache we create a number of points in the circle for it.

You can see the effect of this in the following plot which I produced by simulating storing 10,000 objects in 10 caches using the code described below. On the x-axis is the number of replicas of cache points (with a logarithmic scale). When it is small, we see that the distribution of objects across caches is unbalanced, since the standard deviation as a percentage of the mean number of objects per cache (on the y-axis, also logarithmic) is high. As the number of replicas increases the distribution of objects becomes more balanced. This experiment shows that a figure of one or two hundred replicas achieves an acceptable balance (a standard deviation that is roughly between 5% and 10% of the mean).

Implementation

For completeness here is a simple implementation in Java. In order for consistent hashing to be effective it is important to have a hash function that mixes well. Most implementations of Object's hashCode do not mix well - for example, they typically produce a restricted number of small integer values - so we have a HashFunction interface to allow a custom hash function to be used. MD5 hashes are recommended here.

import java.util.Collection;
import java.util.SortedMap;
import java.util.TreeMap;

public class ConsistentHash<T> {

 private final HashFunction hashFunction;
 private final int numberOfReplicas;
 private final SortedMap<Integer, T> circle = new TreeMap<Integer, T>();

 public ConsistentHash(HashFunction hashFunction, int numberOfReplicas,
     Collection<T> nodes) {
   this.hashFunction = hashFunction;
   this.numberOfReplicas = numberOfReplicas;

   for (T node : nodes) {
     add(node);
   }
 }

 public void add(T node) {
   for (int i = 0; i < numberOfReplicas; i++) {
     circle.put(hashFunction.hash(node.toString() + i), node);
   }
 }

 public void remove(T node) {
   for (int i = 0; i < numberOfReplicas; i++) {
     circle.remove(hashFunction.hash(node.toString() + i));
   }
 }

 public T get(Object key) {
   if (circle.isEmpty()) {
     return null;
   }
   int hash = hashFunction.hash(key);
   if (!circle.containsKey(hash)) {
     SortedMap<Integer, T> tailMap = circle.tailMap(hash);
     hash = tailMap.isEmpty() ? circle.firstKey() : tailMap.firstKey();
   }
   return circle.get(hash);
 }

}

The circle is represented as a sorted map of integers, which represent the hash values, to caches (of type T here).
When a ConsistentHash object is created each node is added to the circle map a number of times (controlled by numberOfReplicas). The location of each replica is chosen by hashing the node's name along with a numerical suffix, and the node is stored at each of these points in the map.

To find a node for an object (the get method), the hash value of the object is used to look in the map. Most of the time there will not be a node stored at this hash value (since the hash value space is typically much larger than the number of nodes, even with replicas), so the next node is found by looking for the first key in the tail map. If the tail map is empty then we wrap around the circle by getting the first key in the circle.

Usage

So how can you use consistent hashing? You are most likely to meet it in a library, rather than having to code it yourself. For example, as mentioned above, memcached, a distributed memory object caching system, now has clients that support consistent hashing. Last.fm'sketama by Richard Jones was the first, and there is now a Java implementation by Dustin Sallings (which inspired my simplified demonstration implementation above). It is interesting to note that it is only the client that needs to implement the consistent hashing algorithm - the memcached server is unchanged. Other systems that employ consistent hashing include Chord, which is a distributed hash table implementation, and Amazon's Dynamo, which is a key-value store (not available outside Amazon).