学步园

*&———————————————————————*
*& Report  ZTEST_PERFORMANCE_1
*&
*&———————————————————————*
*& For Performance Tunning
*&
*&———————————————————————*

REPORT ZTEST_PERFORMANCE_1.

data:
ls_scarr type scarr,
lt_scarr type table of scarr,
ls_spfli type spfli,
lt_spfli type table of spfli,
ls_sflight type sflight,
lt_sflight type table of sflight.

data:
lw_num1 type i,
lw_num2 type i,
lw_num3 type i.

data:
lw_tmp type i.

* main program
start-of-selection.
WRITE:/ sy-UZEIT.

*1. retrieve data from scarr
select * from scarr into table lt_scarr.
describe table lt_scarr lines lw_num1.

DO 100 TIMES.

loop at lt_scarr into ls_scarr.
“2.retrieve data from spfli
select * from spfli into table lt_spfli
where CARRID = ls_scarr-carrid.
lw_num2 = lw_num2 + 1.

loop at lt_spfli into ls_spfli.
“write:/ ’Carrid:’, ls_spfli-carrid,
“        ’Connid:’, ls_spfli-connid.
“uline.
“3. retrieve data from sflight
select * from sflight into ls_sflight
where CARRID = ls_spfli-carrid and
connid = ls_spfli-connid.
“ write:/ ’Carrid:’, ls_sflight-carrid,
“       ’Connid:’, ls_sflight-connid,
“      ’FLDATE:’, ls_sflight-FLDATE.
lw_num3 = lw_num3 + 1.
endselect.
endloop.

endloop.

ENDDO.

write:/ ‘Scarr Records:’, lw_num1,
/ ’spfli Records:’, lw_num2,
/ ’sflight Records:’, lw_num3.

|- CPU time > 40%(Response time - Dispatcher wait data) ==>Se30： Runtime Analysis

• 引起该结果的是由于一些ABAP process，例如Loop等

|- Database time > 40%(Response time - Dispatcher wait data) ==>St05: SQL Trace

• 引起该结果的是由于大量的直接操作数据据等

(1) 设定需要Trace的数据范围

决定将要分析哪些object哪些data哪些process等等.同时很重要的是，决定最后分析结果的“粒度”(Type of aggregation: Full aggregation, aggregation per calling position, no aggregaion).

这些是通过设置SE30运行时的variant来实现的。

(2) Trace 跟踪

现在使用SE30可以直接从SE30进入到所经Trace的Transaction/Program中(同一个user session)，也可以在另外的user session中通过menu中的active runtime analyze等方法来trace跟踪.

(3) 分析结果

分析SE30的结果时，我们一般采取的是“渐进”的方式：先使用full aggregation可以得到程序的主要的FMs或者逻辑处理单元；然后再细化一下分析粒度，aggregation per calling position，甚至no aggregation，这时就可以得到最详尽的统计分析信息: table list, hierarchy等等。

按纽 “Switch on/off” of “In parallel session”

将会有一个screen appears listing all r/3 work processes running on the current application server. To switch on runtime analysis for the object,选中它然edit==>measurement==>switch on/off

Hit list

A list of all called program parts or instructions, sorted according to gross time.It contains one row for each program part,包括name of the program part, gross time, net time,以及number of times this part was called.

Group hit list

A list of individual calls of instructions grouped according to instruction categories(如forms, FM, methods)，包括gross run time and the number of times the instruction was called.

Table hit list

A list showing accesses to DB tables,包括access name, number of accesses, table type,buffering mode以及一些其他信息

Hit list of internal tables

A list of all internal table accesses that indicates the internal name, the number of accesses, the gross and net time taken(absolute value and percentage), the memory required, and other info.

Call hierarchy

A list showing the chronological sequence of analyzed objects, indicating the gross and net time taken and the call level.

hh:mm:ss.ms

time stamp of begin of sending the database request

Duration

操作的总时间in microseconds

Progrm

运行的程序

Object

name of table or view, in the SQL statement

Oper

database operation to be executed

Curs

Name of the databse cursor

ArrSz

size of packets containing the data records sent from DB server to Appliation server

Rcc

number of data records sent by Database operation

RC

return code of db system

Statement

text of the SQL statement

• Structure-identical SQL:它是指SQL结构是一样的，而用于检索的variable的值是不一样的。
• Value-identical SQL:它是指SQL结构、用于检索的variable的值都是一样的。

举个简单的例子:

如SQL 1： select * from <table1> where F1 = <variable1_1>

SQL 2： select * from <table1> where F1 = <variable1_2>

那么，如果这里的<variable1_1> 与<variable1_2>不相等， 则为structure-identical SQL;如果相等，则为value-identical SQL.

从这里可以看出，一般出现structure-identical是在loop中套select，所以这里候应该思考是否可以先一次从DB读取完，然后再在LOOP中使用read 内表。

而value-identical的sql应该使用诸如内表等，坚决避免。