//复制自MSDN，其他答案中没有明确提到非聚类索引的第二点。

集群

Clustered indexes sort and store the data rows in the table or view based on their key values. These are the columns included in the index definition. There can be only one clustered index per table, because the data rows themselves can be stored in only one order. The only time the data rows in a table are stored in sorted order is when the table contains a clustered index. When a table has a clustered index, the table is called a clustered table. If a table has no clustered index, its data rows are stored in an unordered structure called a heap.

非聚集

Nonclustered indexes have a structure separate from the data rows. A nonclustered index contains the nonclustered index key values and each key value entry has a pointer to the data row that contains the key value. The pointer from an index row in a nonclustered index to a data row is called a row locator. The structure of the row locator depends on whether the data pages are stored in a heap or a clustered table. For a heap, a row locator is a pointer to the row. For a clustered table, the row locator is the clustered index key.

聚类基本上意味着数据在表中的物理顺序。这就是为什么每个表只能有一个。

非聚集意味着它“只是”一个逻辑顺序。

群集索引对磁盘上的数据进行物理排序。这意味着索引不需要额外的数据，但只能有一个聚集索引(显然)。使用聚集索引访问数据是最快的。

All other indexes must be non-clustered. A non-clustered index has a duplicate of the data from the indexed columns kept ordered together with pointers to the actual data rows (pointers to the clustered index if there is one). This means that accessing data through a non-clustered index has to go through an extra layer of indirection. However if you select only the data that's available in the indexed columns you can get the data back directly from the duplicated index data (that's why it's a good idea to SELECT only the columns that you need and not use *)

//复制自MSDN，其他答案中没有明确提到非聚类索引的第二点。

集群

Clustered indexes sort and store the data rows in the table or view based on their key values. These are the columns included in the index definition. There can be only one clustered index per table, because the data rows themselves can be stored in only one order. The only time the data rows in a table are stored in sorted order is when the table contains a clustered index. When a table has a clustered index, the table is called a clustered table. If a table has no clustered index, its data rows are stored in an unordered structure called a heap.

非聚集

Nonclustered indexes have a structure separate from the data rows. A nonclustered index contains the nonclustered index key values and each key value entry has a pointer to the data row that contains the key value. The pointer from an index row in a nonclustered index to a data row is called a row locator. The structure of the row locator depends on whether the data pages are stored in a heap or a clustered table. For a heap, a row locator is a pointer to the row. For a clustered table, the row locator is the clustered index key.

聚集索引

一个表只能有一个聚集索引。 通常在主键上进行。 聚集索引的叶节点包含数据页。

非聚簇索引

一个表只能有249个非聚集索引(直到sql version 2005的后续版本支持多达999个非聚集索引)。 通常在任意键上。 非聚集索引的叶节点不包含数据页。相反，叶节点包含索引行。

An indexed database has two parts: a set of physical records, which are arranged in some arbitrary order, and a set of indexes which identify the sequence in which records should be read to yield a result sorted by some criterion. If there is no correlation between the physical arrangement and the index, then reading out all the records in order may require making lots of independent single-record read operations. Because a database may be able to read dozens of consecutive records in less time than it would take to read two non-consecutive records, performance may be improved if records which are consecutive in the index are also stored consecutively on disk. Specifying that an index is clustered will cause the database to make some effort (different databases differ as to how much) to arrange things so that groups of records which are consecutive in the index will be consecutive on disk.

For example, if one were to start with an empty non-clustered database and add 10,000 records in random sequence, the records would likely be added at the end in the order they were added. Reading out the database in order by the index would require 10,000 one-record reads. If one were to use a clustered database, however, the system might check when adding each record whether the previous record was stored by itself; if it found that to be the case, it might write that record with the new one at the end of the database. It could then look at the physical record before the slots where the moved records used to reside and see if the record that followed that was stored by itself. If it found that to be the case, it could move that record to that spot. Using this sort of approach would cause many records to be grouped together in pairs, thus potentially nearly doubling sequential read speed.

In reality, clustered databases use more sophisticated algorithms than this. A key thing to note, though, is that there is a tradeoff between the time required to update the database and the time required to read it sequentially. Maintaining a clustered database will significantly increase the amount of work required to add, remove, or update records in any way that would affect the sorting sequence. If the database will be read sequentially much more often than it will be updated, clustering can be a big win. If it will be updated often but seldom read out in sequence, clustering can be a big performance drain, especially if the sequence in which items are added to the database is independent of their sort order with regard to the clustered index.