-1

用内存优化表替换了一个大表
对于某些东西，我对一些它死掉的东西有很好的响应时间
这是一个复合主键
我可以让它使用主键的唯一方法是搜索特定的行（整个 PK)
不会使用 PK 进行排序或仅使用复合键的一个组件
从现有数据中调整哈希桶的大小

将此链接中的语法用于复合主键
Hekaton： Composite Primary Key in create table statement
CREATE TABLE (SQL Server)

CREATE TABLE [dbo].[FTSindex]
(
    [sID] [int] NOT NULL,
    [wordPos] [int] NOT NULL,
    [wordID] [int] NOT NULL,
    [charPos] [int] NOT NULL,

INDEX [ix_wordID_MO_2] NONCLUSTERED HASH 
(
    [wordID]
)WITH ( BUCKET_COUNT = 524288),
CONSTRAINT [pk_FTSindexMO_2] PRIMARY KEY NONCLUSTERED HASH 
(
    [sID],
    [wordPos]
)WITH ( BUCKET_COUNT = 268435456)
)WITH ( MEMORY_OPTIMIZED = ON , DURABILITY = SCHEMA_ONLY )

select top 10 * from [FTSindex] where [sID] = 100
-- runs in 0 seconds 
-- Index Seek on ix_wordID_MO_2 
-- it is NOT using the PRIMARY KEY pk_FTSindexMO_2

select top 10 * from [FTSindex] where [wordPos] = 100
-- never finishes (I only waited 10 minutes) 
-- will not even display an execution plan

select top 10 * from [FTSindex] where [sID] = 100 and [wordPos] < 1000
-- never finishes (I only waited 10 minutes) 
-- will not even display an execution plan


select top 10 * from [FTSindex] order by [sID]
-- never finishes (I only waited 10 minutes) 
-- query plan is Table Scan 

select top 10 * from [FTSindex] order by [sID], [wordPos]
-- never finishes (I only waited 10 minutes) 
-- will not even display an execution plan

select top 10 * from [FTSindex] where [wordID] = 100 and [sID] = 856515 
-- runs in 0 seconds
-- Index Seek on ix_wordID_MO_2

select top 10 * from [FTSindex] where [wordID] = 100 and [sID] = 856515  and [wordPos] < 1000
-- never finishes (I only waited 10 minutes) 
-- will not even display an execution plan

select * from [FTSindex] where [sID] = 100  
-- 45 seconds to return 1500 rows 
-- table scan 

select * from [FTSindex] where [sID] = 100  and [wordPos] = 1133 
-- runs in 0 seconds
-- this uses the pk_FTSindexMO_2 
-- this is the only way I could get it to use the primary key

注意原始（非内存优化表）
所有这些查询都在 0 秒内运行
我并不是说每个
都在 0 秒内运行

我认为这总结了我的问题
Troubleshooting Common Performance Problems with Memory-Optimized Hash Indexes

不使用 HASH 作为主键似乎已修复它 

CREATE TABLE [dbo].[FTSindex]
(
    [sID] [int] NOT NULL,
    [wordPos] [int] NOT NULL,
    [wordID] [int] NOT NULL,
    [charPos] [int] NOT NULL,

INDEX [ix_wordID_MO_2] NONCLUSTERED HASH 
(
    [wordID]
)WITH ( BUCKET_COUNT = 524288),
CONSTRAINT [pk_FTSindexMO_2] PRIMARY KEY NONCLUSTERED 
(
    [sID] ASC,
    [wordPos] ASC
)
)WITH ( MEMORY_OPTIMIZED = ON , DURABILITY = SCHEMA_ONLY )

最后注意我回到旧的基于磁盘的表
在应用程序使用的实际查询内存优化较慢优化
的内存加载速度更快，但该表是一次写入和多次读取

4

1 回答 1

2

哈希索引不适用于范围扫描。范围索引是。

于 2014-05-13T21:43:54.523 回答