Data Warehouse Hardware

1. Disk I/O, 硬盘IO速度

　　硬盘的IO速度一直都是数据库的瓶颈，所以有条件的情况下尽可能的使用高IO的磁盘。

　　可以使用微软的工具SQLIO测试磁盘的IOPS

2. CPU的主频，

　　DW和传统的OLTP数据库在使用场景上不一样。

　　传统的OLTP数据库具有[事务小][并发多]的特点；而DW的数据库相比较具有[事务大][并发少]的特点。

　　所以对比起来，传统的OLTP数据库可以使用[低主频][多核]的硬件架构，而DW建议使用[高主频][少核]方案。

　　上述都是相对情况，对于不差钱的土豪，高主频，多核当然是更好的选择。

我们可以计算的是要满足具体的业务需求，需要多少CPU(Core)，多少内存。

MCR，Maximum Consumption Rate，这是一个Core的吞吐量指标

3. 计算MCR

　　可以使用下面的脚本计算出当前计算机的MCR

　　

USE master;

-- Create a database for benchmark queries
IF EXISTS (SELECT * FROM sys.sysdatabases WHERE name = 'BenchmarkDB')
DROP DATABASE BenchMarkDB;
GO
CREATE DATABASE BenchMarkDB;
GO
USE BenchMarkDB;

-- Include a heap and a table with a clustered index
CREATE TABLE heap_table
(col1 integer identity,
 col2 integer,
 col3 varchar(50));

 CREATE TABLE clust_table
(col1 integer identity PRIMARY KEY CLUSTERED,
 col2 integer,
 col3 varchar(50));

-- Insert 100 rows to start with
DECLARE @i integer = 0;
WHILE @i < 101
BEGIN
   SET @i = @i + 1
   INSERT INTO heap_table VALUES (@i, CAST(@i%5 AS varchar))
   INSERT INTO clust_table VALUES (@i, CAST(@i%5 AS varchar))
END;

-- Now keep reinserting exponentially until the tables each contain 2 million rows
WHILE (SELECT COUNT(*) FROM clust_table) < 2000000
BEGIN
 INSERT INTO heap_table
 SELECT col2, col3 FROM clust_table;
 INSERT INTO clust_table
 SELECT col2, col3 FROM clust_table;
END;

USE BenchmarkDB
GO

SELECT SUM(Col2) FROM heap_table WHERE col1 % 3 = 1
GROUP BY col3;

SELECT SUM(Col2) FROM clust_table WHERE col1 % 3 = 1
GROUP BY col3;


SET STATISTICS IO ON;
SET STATISTICS TIME ON;


-- run these muliple times and take an average of the logical reads and CPU time
SELECT SUM(Col2) FROM heap_table WHERE col1 % 3 = 1
GROUP BY col3
OPTION (MAXDOP 1);


SELECT SUM(Col2) FROM clust_table WHERE col1 % 3 = 1
GROUP BY col3
OPTION (MAXDOP 1);

/* Max Consumption Rate (MCR) is the average of (logical reads / CPU time in seconds) * 8 / 1024
  (or put another way, the size of the table in MB / CPU time in seconds)
  This gives us the throughput of a core

  To estimate the no. of cores required, use the following formula:
  (Amount of data scanned in an average query / MCR) * Concurrent Sessions / Target response time
  For example:
 (18000 MB/200 MBs) * 10 users / 60s response time = 15 cores (round up to 16)   */

4. Memory内存需求

　　最少1核对应4G内存，或者对每组CPU给64-128G内存