原文是E文，做一下翻譯并且提取一些重要的思路。

1.定義業(yè)務(wù)規(guī)格

• What are your most critical business processes and how often are they processed (e.g. number of sales activities per day, number of client requests per day, etc.)
• 最重要的業(yè)務(wù)流程是什么，使用頻率如何
• What is regarded (required) as acceptable response times from the online application? How about reports?
• 可以接受的在線(xiàn)應(yīng)用程序的響應(yīng)時(shí)間是多少，報(bào)告又如何
• How much data must be kept available on the live database (over 1 month, 1 year, 5 years?). Can data be archived? How and when?
• 數(shù)據(jù)在live數(shù)據(jù)庫(kù)中要保存多久，可不可以歸檔，如何歸檔，什么時(shí)候歸檔
• How many concurrent application users are expected?
• 并發(fā)的用戶(hù)數(shù)是多少
• Are there periods of more than average usage? (E.g. Daily between 9:00 A.M. and 10:00 A.M. /yearly ’round about Christmas.) What business processes will be impacted the most?
• 業(yè)務(wù)的高峰期在什么時(shí)候，那些流程會(huì)被影響
• How much data latency is acceptable for reports?
• 報(bào)表數(shù)據(jù)延遲，多少時(shí)間內(nèi)可以接受
• Etc.
• 等等

2.分析業(yè)務(wù)

• How many inserts/updates/deletes/selects will be performed on the most important database objects when the database is averagely stressed? How much when there is a period of more than average usage?
• 在平均負(fù)載的情況下，在重要的數(shù)據(jù)庫(kù)對(duì)象中，有多少insert，delete，select會(huì)發(fā)生。在業(yè)務(wù)高峰期有是多少
• How large will your database core objects become (over 1 month, 1 year, 5 years)?
• 若干時(shí)間后，數(shù)據(jù)庫(kù)的核心對(duì)象會(huì)變得多大？
• How many concurrent database users are expected? (This is not equal to application users if connection pooling is implemented.)
• 數(shù)據(jù)庫(kù)的并發(fā)量是多大
• What should be regarded as acceptable database response times? For online requests/for reporting requests?
• 可以接受的數(shù)據(jù)庫(kù)響應(yīng)時(shí)間是多少

3.定義磁盤(pán)子系統(tǒng)的需求

• 可以使用sqlio 來(lái)測(cè)試 磁盤(pán)子系統(tǒng)
• 也可以使用 SQLIOStress
• 如何達(dá)到標(biāo)準(zhǔn)文章并沒(méi)有提及，可以根據(jù)業(yè)務(wù)繁忙時(shí)段，io的吞吐量也確定，多少才能滿(mǎn)足要求

4.為準(zhǔn)備腳本

• Ask a business user to perform a realistic work process as he would in real life situation, including realistic intervals between tasks.
• 咨詢(xún)終端用戶(hù)實(shí)際工作流程，真實(shí)的工作狀況，包括任務(wù)之間的時(shí)間間隔
• Trace this activity using SQLProfiler (only? textdata ?and? starttime ?are important)
• 使用SQLProfiler跟蹤操作
• Replace variables with placeholders for parameters that can be defined at random. (This is where you wanted all database interactions to be performed using stored procedures.)
• 使用隨機(jī)數(shù)來(lái)代替用戶(hù)操作的變量
• Add think times to the script.
• 增加延遲
• 相關(guān)的操作作者給出了代碼，可以參考原文

5.添加虛擬用戶(hù)

• 相關(guān)的操作作者給出了代碼，可以參考原文

6.添加測(cè)試數(shù)據(jù)

• 可以使用DTS或者SSIS初始化數(shù)據(jù)

7.監(jiān)控

I cover this topic in my previous article on? Monitoring .

8.迭代測(cè)試

It is important to run multiple iterations using different parameter combinations but you also need to perform multiple iterations of the same scenario to avoid accidental external impact.

　　使用不同的參數(shù)組合迭代測(cè)試十分重要，你也需要執(zhí)行在同一個(gè)場(chǎng)景下執(zhí)行迭代測(cè)試來(lái)避免意外情況的影響。

原文： SQL Server Monitoring in 8 Steps: Lessons From the Field ?- 02 Aug 2005

數(shù)據(jù)庫(kù)壓力測(cè)試思路