SAP Architecture

R/3 Workload Analysis
Deming Cha
SAP Technology Inc.
SAP America, Inc. prf3.ppt December, 1996
Workload Analysis
R/3 Architecture Review
Dialog and update processing
R/3 Performance Monitoring Tools

The roadmap to approach performance problems
Expensive SQL statements

As a diagnosis example
Memory Management for R/3 3.x

Small vs. big configurations
ABAP/4 trace tools
Page 1
Workload Analysis
R/3 Architecture Review
R/3 System Review - Dialog Processing

1
Network
Wait Queue (Memory)
TCP/IP sockets
Presentation Layer
2 3 4
Dispatcher
13
14
Application Layer
Roll buffer Page buffer Page file Roll file Network
Work Process
Roll mem Page mem
...
DB Interface
5 12 11
SAP Shared Pool Buffers
6 10
Database process Database Buffers

9 7 8
Database Layer
Page 2
R/3 System Review - Update Processing

Order form Order form
Presentation Layer
Sold to
Mr. Jack Knife
Sold to
Mr. Jack Knife
Check availability
Confirm sale
Check availability
Confirm sale
Item #
00086
Desc.
BandAids
Quantity
10
Item #
00086
Desc.
BandAids
Quantity
10
Network
6 Send DB COMMIT Dialog work is finished 5
Dialog Process
Order Entry ABAP Code
Dispatcher
Enqueue Process
Lock Material #00086
7
Update Process
Update Code
Application Layer
Send DB COMMIT Update work is finished 10 9
11
Network
Database
vblog vbak,vbap,vbuk,vbup,vbpa,vbep,vbkd,vakpa,vepvg,koclu
The R/3 Performance Timing Statistics

CPU Time
Roll in DB Time Network
Roll out
Load Time
Process Time
Wait Time
Dispatch Time Response Time
Database Server
Application Server
Presentation Server
Page 3
Workload Analysis
R/3 Performance Monitoring Tools
Main Monitors for Analysis
Workload Monitors
Database Monitor
st04
SAP Buffers Monitor

st02
Process and OS Monitors

sm50, st06
Transaction Monitor
st03
Page 4
The Workload Monitor (st03)

Performance: Workload AnalysisWorkload for Server hs1036
Other period ... Top time Time profile Transaction profile Memory profile Accounting profile
SAP application server

SAP-System Server Date 4795 E11 hs1036 27.07.1994
Workload Summary of an Instance

First record Last record Elapsed time 07:33:35 14:20:30 06:41:55
Response times for all task types

CPU time Elapsed time Dialog steps Av. Response time Av. CPU time Av. Wait time Av. Load time Av. DB requests Av. bytes request Roll-ins Roll-outs 731,2 24115,0 s s Data base calls Database requests Dir. reads Seq. reads Changes Time per DB requests Dir. reads Seq. reads Changes Av. time / Roll-in Av. time / Roll-outs 33.422 87.255 22.198 64.313 744 20,3 6,9 21,2 334,8 ms ms ms ms
4.950 885,1 ms 147,7 ms 5,3 ms 6,4 ms 357,0 ms 35,6 kB 6.284 4.784
3,1 ms 10,0 ms
Task types
Current Total Other Dialog Background

Workload: Transaction Profile Report
Monitoring
Choose >> Sort
System
Text
Help
Transaction Profile Information

DB time total(s) avg(ms)
Program or Dialog Time Steps
Response time CPU time Wait time total(s) avg(ms) total(s) avg(ms) total(s) avg (ms)
*TOTAL* VA01 MainMenu SE37 VTAA VA41 ME21 CU04 VA02 VOV4 System VOV7
488 234 80 39 36 33 20 13 5 4 4 4
520 266 10 55 45 56 59 10 4 4 1 1
1.065 1.138 124 1.418 1.237 1.705 2.938 765 876 898 303 341
261 145 7 15 28 38 14 2 2 2 1 1
534 620 91 395 765 1.144 703 169 358 483 248 188
4 1 0 0 0 0 1 0 0 0 0 0
8 6 5 6 5 5 71 5 5 5 4 5
258 121 2 40 16 19 45 8 3 2 0 1
530 518 26 1.037 450 562 2.242 627 530 413 11 158
Page 5

Workload: Top 40 with Respect to Time Consumption
System 47 K11 Server hs1036 Analyzed day We 27.07.1994 until 14:20 Sorted by Time of day
Top 40 dialog steps by response time ( out of 4950 ) End time Tcode Program 11:07:03 12:54:24 09:11:10 13:43:58 13:51:01 11:29:50 13:49:47 11:14:30 14:12:46 13:53:03 11:22:44 12:55:17 11:36:12 13:40:08 12:48:43 13:53:02 11:12:56 DI03 C001 C0R2 AM02 SQ01 0PJK SE01 C0R2 ST03 AM02 C0R2 C001 ST03 C0R2 C002 SQ01 C0R2 T D D D D D D D D D D D D D D D D D Scr. 0120 0010 5101 0120 1000 0120 0120 0300 0010 0100 0120 0100 0010 0900 0100 0120 5101 Wp User 0 0 2 1 2 0 0 2 0 1 0 0 0 0 0 2 2 -?-?-?-?-?-?-?-?-?-?-?-?-?-?-?-?-?-
Top Time Profile

Response Dispatch Wait CPU DB req. Load/Gen time (ms) time (ms) time (ms) time (ms) time (ms) time (ms) 165.401 97.317 95.718 86.984 86.520 67.728 63.650 58.828 55.822 53.393 52.924 45.741 44.824 41.059 39.715 39.380 37.420 165.400 97.316 95.717 86.983 86.519 67.725 63.646 58.467 55.821 53.392 52.923 45.740 44.823 41.058 39.713 39.379 37.427 1 1 1 1 1 3 4 361 1 1 1 1 1 1 2 1 1 610 18.730 6.930 1.440 1.600 6.170 90 10.820 1.040 3.180 1.780 2.030 1.120 15.650 3.470 2.300 8.050 163.016 20.763 10.840 86.437 5.677 5.546 63.404 9.015 51.521 786 49.673 10.171 41.748 275 26.739 4.089 3.213 1.269 33.836 73.388 77 21.558 38.613 21 19.274 3.230 52.190 1.057 19.539 1.821 40.278 6.184 1 18.985
RDN03S01 SAPLC0K0 SAPLC0K0 SAPMA01M !Q0100BA SAPMSVMA RDDIT070 SAPLC0K0 SAPMSS03 SAPMA01M SAPLC0K0 SAPLC0K0 SAPMSS03 SAPLC0K0 SAPLC0K0 !Q0I00BA SAPLC0K0
Process Overview Monitor (sm50)

Table Statistics Workload: Tables for Application Server hs1036
.Ty. PID DIA 191 DIA 221 DIA 210 DIA 205 DIA 150 DIA 178 DIA 203 Status Running Waiting Waiting Waiting Waiting Waiting Waiting ReasonStart Yes Yes Yes Yes Yes Yes Yes Err Sem CPU Time Program Cl. User RSMON000 002 GOETHALS Action Tabl
WP WP WP Application Instance
Maria: Close these boxes off with a line
Resources
Operating System
Page 6
The OS Monitor (st06)

OS Monitor HP-UX
Refresh display Detail analysis menu Collector info
CPU Utilization user % system % idle % System calls/sec Interrupts/sec Memory Physical mem avail kb Pages in/sec Pages out/sec Swap Primary-swap Size swap-space kb kb
41 42 17 3.890 1.562
Count Load average
1 min 5 min 15 min Context switches/sec
1 1,78 1,40 1,31 2.927 700 0 0 3.063.532
524.288 Physical mem free kb 0 Kb paged i/sec 0 Kb paged out/sec 2.052.096 Free swap-space 3.064.904 kb
Disk with highest response time c28d0s0 Response time Name 6 Queue Utilization 0 Avg service time Avg wait time ms 35 Kb transferred/sec
ms ms
680 0 680
OVR 04:50PM
K11(0) hs0013
The Database Monitor (st04)

Data Base Performance Analysis: Oracle Data Base Overview
Refresh display Detail analysis menu
Database K11 Day, Time 28.09.1994 D0:07:05 Database summary DB Server hs0013 Since start up 20.09.1994 D6:33:56 Data buffer Size Quality kb 240.000 % 97 Reads 1.654.230.588 Physical reads 44.627.488 writes 6.080.448 Busy waits 374992 Busy wait time ms 141424070 Log buffer Size kb Entries Allocation retries Alloc fault rate % Recursive calls Parses
K11(0) hs0013
Shared Pool Size kb DD-Cache quality % SQL Area getratio % SQL Area pinratio % Calls User calls commits rollbacks
162.024 95 98 100 60.589.413 586.102 4.153
20.480 42.457.752 2.771 0 23.091.509 2.375.072

OVR 04:47PM
Page 7
The Database Monitor (st04)

Exclusive Waits in the ORACLE Data Base
04.08.1994
16:01:56
Exclusive session-lock situations (no shadow process)
Lock-Holder (Host, Pid) Lock-Waiters (Host, Pid) hs5101 hs5101 16032 16033
Object-Name
T100
...
Host with Lock,PID
Resource Waiting
Database analysis Cho. 02.05.1996

Total Execution 6.746.205 4.971.100 3.413.492 3.002.588
Edit
Data Base Performance: Shared SQL Goto Monitor System Help

?
Reset 16:59:00
Since Reset
s.DBst.
Select table
Sort
More Information
Shared Cursor Cache

Users Open 5 209 230 226 Disk Reads 186.702 874.636 1278.997 562.411
( last reset at 02.05.1996 09:23:30 )

Reads/ Buffer Execution Gets 0,0 26.500.852 0,2 19.939.081 0,4 13.670.450 0,2 8.978.357 Parse Calls 232 6.992 20.096 17.036
Current Execut. 0 0 229 0
Double click on statement Data Base Performance: Shared SQL Double click on statment Database analysis Edit Goto Monitor System Help
? Choose
Total 02.05.1996 Total Execution Execution 4.971.100 3.413.492 3.002.588
Explain
Current 16:59:00
DIC Info
More Inform.
Users Disk Reads/ Buffer Parse Shared Cursor Cache ( last reset at 02.05.1996 09:23:30 ) Current Disk Reads/ Buffer Parse Execute. Users Open Reads Execution Gets Calls Execut. Open Reads Execution Gets Calls 0 209 874.636 0,2 19.939.081 6.992 229 230 1278.997 0,4 13.670.450 20.096 0 226 562.411 0,2 8.978.357 17.036
SELECT TABNAME, BLOCKNR, FIELDSLG, FIELDS FROM "DDNTF" WHERE TABNAME = :A0000 ORDER BY TABNAME, BLOCKNR #
BIN(1)
hs0011
OVR
11:38AM
Page 8
The Setup/Buffers Monitor (st02)

Storage Usage and Requirements
System : ds0201_Q30_86 17:55:29 Tune summary Startup: 28.02.1996 12:59:10 Free space [%] Dir.size Entries Free directory Entries [%] Swaps a D Date & time of snapshot : 28.02.1996 Buffer Hitratio [%] Nametab (NTAB) Table definition Field description Short NTAB Initial records Program CUA Screen Calendar Tables Generic key Single record Roll & Page
Allocated [kB] [%]
94,37 90,38 96,23 89,60 89,55 89,15 94,17 99,32 98,05 76,47 Current use [%] [kB]
1.164 5.621 1.612 2.112 131.097 5.000 2.441 488 15.625 10.000 Max.use [kB] 36.304 672 31.744
154 325 612 51 25.052 4.392 1.867 414 3.662 9.132 [%]
16.45 6.50 61.20 3.40 19.27 90.35 82.39 85.89 24,32 91,69 [kB]
12.288 24.577 24.577 24.577 16.250 2.500 5.000 200 6.000 500
11.952 24.577 24.577 24.577 14.896 736 4.370 144 4.401 375
11.952 24.237 24.405 24.308 13.74 29,44 87,48 72,00 73,35 15,00 Hitratio [%]
0 8.882 0 0 10.062 1.910 2.321 0 0 0 Fr Entries 0 0
In Memory
On disk
SAP cursor cache
Roll area Paging area Extended Memory
14,34 18.792 0,43 560 100,00 31.744
32.000 8.000 31.744
99.072 123.072
IDs Statements
97,40 78,00
Table statistics workload: Tables for application server hs0011 Table statistics Edit Goto Environment Monitor System Help ? Sort Analyze table Since startup Reset Previous day Next day Overview <-> Detail
Table access summary sorted by
DB requests ABAP/4 PROCESSORS REQUESTS
Table *Total* NTAB DDNTF NTAB DDNTT ATAB TCP01 SNAPT MONI ATAB TCP02 TRPUT ATAB TFO04 VBHDR ATAB TCP07 ATAB TSE02
Type
Buffer
Total 468.861
Seq. 166.026 922 0 39.630 20.913 7.110 8.509 8.018 6.902 5.952 5.948 5.774 BIN(1)
Dir. 295.538 198.436 59.436 0 0 0 0 0 0 23 0 0 hs0011
Changes 7.297 1 3 0 34 1.821 0 0 0 154 0 0 11:38AM
System System Pool Table Table Pool Table Pool Table Pool Pool full
199.368 59.439 39.630 20.947 8.931 8.509 8.018 6.902 6.129 5.948 5.774
OVR
Page 9
SQL Trace (st05)

SAP work process
ABAP Job
DB interface Local buffers
Local Trace file SQL
DB
ABAP Trace - Runtime Analysis (se30)
PROGRAM XYZ000. DATA: A DATA: RESULT(20) A = 123. PERFORM CALL USING A. WRITE: / RESULT. FORM CALL. ... ENDFORM. TYPE P. TYPE C.
Time Duration Stamp ... ... ... ...
Statement Type ... ... ...
Runtime Summary
... ...
Trace file
Page 10
Analyze the problem

(Average) Performance problem ? Yes All Users Affected ? No Which is bad ? CPU time
Check Workload on all Servers
CPU time << Process time (Continued) Wait time
ABAP or SQL Trace
Load time DB request time
Analyze the problem (Continued)

(Average) CPU time
Amount of time occupying the work process
CPU time << Process time
I/O Bottleneck Network problems, CPU bottlenecks
Wait time
Configuration problem - Not enough work processes Long running transactions - All work processes busy Configuration problem - R/3 Buffers too Small Installation problem - Missing indices DB Server CPU has heavy workload More memory in DB server needed Too many disk sorts - Missing Indices Too many extents exist on a table
Load time DB request time
Page 11
Performance Problem Roadmap (1)

If Top Time MAIN_MENU stat shows response times reasonable SQL Trace on TOP TIME Transactions ABAP Trace on TOP TIME Transactions
Specific Problem
Inspect overall Workload statistics for problem areas

General Problem
Check if R/3 buffers need adjustments Are DB buffer qualities< 97% Check missing indexes Use individual transaction workload stats, identifying intensive transactions.
Check if objects can be buffered Allocate more data Buffers
If Av DB time > 40% Response Time
Consider tuning DB in order of highest gains: App. SQL Tuning 70% Memory 15% Disk I/O 10% Memory Contention 5% Build new index(s) when necessary Split instance into 2 smaller instances, separating roll mechanisms
Use SQL trace on transaction
If Av Roll > 2x Rated disk access time
Increase in memory roll area
Offload roll file to faster or more dedicated disk
Performance Problem Roadmap (2)

If Av Wait Time > 1% Response Time Problem Occurring Now Use CPU button of Process Monitor Identify long running Jobs Problem occurred in Past Use Top Times/Tcode WL Monitors, Reference problematic transactions Tune long running Jobs Increase number of needed Workprocesses If Semaphore 6 is continually present, investigate resource conflict situation
If Av CPU Time > 40% Response Time
Use ABAP Trace on time consuming transactions Check R/3 buffers and indexes Analyze disk/network thruputs using OS monitor and/or Network Monitor If Page-Outs exist analyze Extended Memory allocations first If the Extended Memory Allocations: CURRENT USE=IN MEMORY
Analyze efficiency of coding logic and reprogram
If Av Load Time > 10% Response Time
If Av Process Time << Av Response Time
Consider buying more memory or reduce allocations on the OS Jump to Extended Memory Solution Roadmap Faster CPU needed or introduce additional R/3 app server
If Load stats > 3 Check for processes using CPU and remove SAP America, Inc. prf3.ppt December, 1996
If load persists tune all CPU intensive R/3 apps
Page 12
Extended Memory Solution Roadmap

Preliminaries Let W = number of workprocesses X = size of additional available memory Y = min(ztty/roll_area * 10% , X/W)
Jump to Extended Memory Solution Roadmap
No Allocate All available memory X to em/initial_size More than 10% of users use heap memory
Yes Allocate an additional Y MB to ztty/roll_area and the remainder of X-(Y*W) to em/initial_size
Recheck heap statistics and reallocate memory from ztty/roll_area to em/initial_size to conform to the 10% rule if necessary
Workload Analysis
Expensive Select Statement
Page 13
Goals
What is an expensive SELECT statement? Which consequences has an expensive SELECT statement? How to detect an expensive SELECT statement?
Using the Statistical Records Using the SQL Trace Using the Shared Cursor Cache Using the Process Overview Using the Where used list
How to tune an expensive SELECT statement?

Creating an Index Understanding the DB optimizer Rewriting inefficient coding
What is an expensive SELECT Statement?

Long response time Many data blocks are scanned to find the selected records Long DB request time caused by too many buffer gets
Page 14
Buffer gets
Application Server A
ABAP/4 Program
SELECT * FROM T001 WHERE ...
R/3 Database Interface records
Table Buffer
DBMS
DBMS Processes
Database Buffer buffer gets
Database
Which consequences has an inefficient SELECT statement?

Database is busy reading many data blocks High CPU load on database server Workprocess is blocked by report
high wait times for other processes
Many blocks are displaced from database buffer

Cache hit rate for other SQL statements suffers
System wide performance can be harmed by expensive statements
Page 15
How to detect an expensive SELECT statement?

Find out
reports/transactions where the database request time is a large fraction of response time SELECT (SQL) statements with a high number of buffer gets
For these SELECT statements find out

the name of the table the where clause The indices that are used
Sometimes you have to look at the coding. In this case you need:
The name of the report or transaction containing the expensive statement
Which of the information is provided by the different screens?

Statistical Records
ST03
SQL Trace
Oracle Process Shared Cursor Overview Cache

ST04Detail analysis Menu SQL Request table name index used SM50
Where used list

SE12:Utilities Where used list report / transaction name table name required sometimes a table is used in many different places
ST05
report / transaction
table name index used
table name report / transaction name Critical transaction must run at the same time
report / transaction name or user name required
Oracle specific
Page 16
Statistical Records: Sample Screen

?
Program or Tcode
Response time CPU time avg (ms) avg (ms)
Wait time avg (ms)
DB time Requested avg (ms) kBytes
VF01 VF01 VF01
52.345 57.278 54.987
2.987 2.988 2.956
1 2 1
49.345 48.536 47.359
1.523 1.523 1.523
Statistical Records: How to interpret

Performance problem ? Load time
Wait time
All Users Affected ?
Yes
R/3 on all Servers CPU time << Process time
A set of expensive statement can block wps and exploit resources
Depending on the times of executions all users could be affected. Involved reports have a high DB request time.
DB request Expensive SQL Statements time many buffer gets
Page 17
SQL Trace: How it works

Application Server A
ABAP/4 Program
SELECT * FROM T001 WHERE ...
R/3 Database Interface SQL trace SQL records
Table Buffer
DBMS
DBMS Processes
Database Buffer
Database
SQL Trace: Sample Screen

? Detail DDIC info Explain SQL
PID= 0000017643 Client= 900 User= WRKL0 T Table Database-Request REOPEN 149 SELECT WHERE "MANDT" = 900 AND "POSNR" = 0010 FETCH 149 Records: 1 Returncode: +0
Transaction/Report= VA01 hh:mm:ss.ms Duration 16:35:52.511
104 D ZXYZ D ZXYZ
16:35:52.512 2.626.051
Page 18
Oracle Shared Cursor Cache: Sample Screen

? Sort Reset Since Reset Since DB Start
Shared Cursor Cache
detail stats .
(last reset at 25.03.1996 15:17:55 )
25.03.1996 16:54:36
Total Execution
... Disk Reads
Reads/ Execution
Buffer Gets
Gets/ Execution
... SQL Text
Statements generated from ABAP/4 have double quotes
57 27 2 149 22
554 0 425 317 13
9,7 0,0 212,5 2,1 0,6
5.200.089 2.200.022 57.901 47.691 29.286
91.229,6 81.482,3 28.950,5 320,1 1.331,2
SELECT "MANDT" , "VBELN" , "POSNR" SELECT "MANDT" , "VBELN" , "POSNR" SELECT TABNAME, TABFORM, REFNAME FR SELECT "TABNAME" , "VARKEY" , "DAT SELECT "TABNAME" , "VARKEY" , "DAT
Process Overview: Sample Screen

? CPU No. 0 1 2 3 4 5 6 7 8 9 10 11 12 Ty. DIA DIA DIA DIA DIA DIA UPD UPD ENQ BTC BTC SPO UP2 Refresh PID 17639 17640 17641 17642 17643 17659 17665 17670 17671 17675 17678 17679 17683 Status running running running waiting running waiting waiting waiting waiting waiting waiting waiting waiting ReasonStart Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Time Program 38 ZZXXYY00 35 ZZXXYY01 32 ZZXXYY01 Cl. 900 900 900 User SAP_PERF SAP_PERF SAP_PERF Action Sequential read Sequential read Sequential read Table ZXYZ ZXYZ ZXYZ
RSMON00 900 WRKL0
Page 19
Where Used List: Sample Screen

?
Program ZFILLIPS
Found locations/short description Fills table ZXYZ with dummy data ZZXXYY00 33 24 tables: zxyz. include structure zxyz. and lgort = int_lips-lgort and matnr in sl_matnr. ZZXXYY01
26 ranges: sl_matnr for zxyz-matnr. 49 select * from zxyz where werks = int_lips-werks
How to tune an expensive SELECT statement?

Expensive statement
DDIC Info Inefficient coding? No Yes
SQL Explain good optimizer decision?
Where used list
a suitable index exists? No
Yes
Yes
Check application re-code
No No "Missing Index"? Yes
create secondary index in DDIC further rules!

create index on DB
re-code or change index
Page 20
Summary
An 'Expensive SQL statement' needs many buffer gets It can cause system wide performance problem Analyzing an expensive SQL statement
Statistical Records: high database request time SQL Trace: statements with long response time Cursor Cache: high number of buffer gets Process Overview: report & table name, no CPU usage while DB access Where used list: table report
Tuning an expensive SQL statement

Creating an index Understanding the DB optimizer Rewriting expensive coding
Workload Analysis
Memory Management for R/3 3.x
Page 21
R/3 old memory management concept (version 2.x)

Work Process Roll Area Paging Area
Access Privileges Entry Data ABAP/4 and Dynpro Management Information Program variables and stacks OCCURS

Work Process Roll Area Paging Area
Extensive Collection of data e.g., report lists

Paging Space
Page 22

Work Process 1 Roll Area Paging Area
Shared Memory Paging Buffer Roll Buffer
Page File
Roll File
Paging Area
Roll Area
Work Process 2
Small configuration
R/3 User
R/3 User
R/3 User
R/3 User
Work Process
Roll Area Page Area
Work Process
Roll Area Page Area
Page File
Roll File
Page 23
Small configuration
Process Program Overview of Processes User session Edit Goto System
? CPU Refresh Delete session Debugging Detail info
Help
No.Ty. 0 1 2 3 4 5 6 7 8 9 10 11 DIA DIA DIA DIA DIA DIA UPD UPD ENQ BTC SPO UP2
PID 2431 2432 2433 2434 2435 2436 2437 2438 2439 2440 2441 2442
Status running running running running running running waiting waiting waiting waiting waiting waiting
Start Err Sem CPU Time Program Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes 1 6 1 1 2 2 SAPDF05X RSMON000 SAPLKAIP SAPMF05A SAPMF05L SAPMF05A
Cl. 900 001 900 900 900 900
User SAP_PERF SAP* SAP_PERF SAP_PERF SAP_PERF SAP_PERF
Action
Small configuration
Workload Edit
Choose >>
Workload: Transaction Profile Report Goto Environment Monitor System Help

? Aggregation... Text
Graphics Sort
Program or Tcode *TOTAL* FB01 FBZ1 FB03 FBL5 SM50 : :
Dialog steps 914 386 244 119 118 21 : :
Response time total(s) avg(ms) 830 492 142 56 118 7 : : 908 1.274 581 468 1.003 334 : :
CPU time total(s) avg(ms) 136 67 32 11 22 1 : : 149 174 130 94 187 62 : :
Page 24
Small configuration
Workload Edit
>> Previous detail
Workload: Task Type Profile Report Goto Environment Monitor System Help
? Next detail Download CPU time total (s) avg (ms) 0 136 0 2 149 34
Task type BUF.SYN DIALOG SPOOL
Dialog steps 8 914 10
Response time total (s) avg (ms) 2 830 0 209 908 49
Summary for tasktype DIALOG Used work processes Dialog steps with DB calls waiting > 50 ms with roll in Avg elapsed time Avg not in workp. Avg response time CPU time wait time load time DB request time Roll time in out
4 914 499 229 908 2.380,8 1.472,4 908,4 149,2 192,6 95,3 46,1 115,3 225,9 ms ms ms ms ms ms ms s s
First record time Last record time Elapsed time Response time CPU time wait time oad time DB request time Roll in out
10:53:36 11:02:40 544,0 830,3 136,3 176,0 87,1 42,1 978 976 s s s s s s
New Memory Management Concept (version 3.x)

Dialog Work Process 1
Roll Seg Process Memory (exclusive per wp) Roll Seg Local Seg
Shared Seg
Shared Shared Seg Seg
Shared Seg
Extended Memory (sharable by all wp's Roll Seg Local Seg
Roll Seg
Process Memory (exclusive per wp)
DialogWork Process 2
Page 25
New Memory Management Concept (version 3.x)

Using traditional Rel 2.2 Paging
Roll Seg Process Memory (exclusive per wp) Paging Area
Extended Memory (sharable by all wp's
Shared Seg
Shared Seg
will stay small because neither Lists nor Internal Tables are contained
Paging space
Parameter abap/use_paging = 1
Internal Tables Lists, etc.
Large configuration
R/3 User
Roll Area Page Area
R/3 User
R/3 User
Roll Area Page Area
R/3 User
Work Process
Work Process
Roll File
Extended Memory
Page File
Local Area
Local Area
Page 26
Large configuration
Process Program Overview of Processes User session Edit Goto System Help
? CPU Refresh Delete session Debugging Detail info
No.Ty. 0 1 2 3 4 5 6 7 8 9 10 11 DIA DIA DIA DIA DIA DIA UPD UPD ENQ BTC SPO UP2
PID 2431 2432 2433 2434 2435 2436 2437 2438 2439 2440 2441 2442
Status running running running running running waiting running waiting waiting waiting waiting waiting
Start Err Sem CPU Time Program Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes 1 1 2 1 1 SAPDF05X RSMON000 SAPLAINT SAPLGLIN SAPDF05X RSM13000
Cl. 900 001 900 900 900 900
User SAP_PERF SAP* SAP_PERF SAP_PERF SAP_PERF SAP_PERF
Action
Large configuration
Workload Edit
Choose >>
Workload: Transaction Profile Report Goto Environment Monitor System Help

? Aggregation... Text
Graphics Sort
Program or Tcode *TOTAL* FB01 FBZ1 FBL5 FB03 SM50 : :
Dialog steps 677 307 169 84 83 12 : :
Response time total(s) avg(ms) 423 492 90 56 19 4 : : 625 789 535 665 227 296 : :
CPU time total(s) avg(ms) 96 52 20 14 7 1 : : 141 168 120 161 82 45 : :
Page 27
Large configuration
Workload Edit
>> Previous detail
Workload: Task Type Profile Report Goto Environment Monitor System Help
? Next detail Download CPU time total (s) avg (ms) 0 96 0 0 141 34
Task type BUF.SYN DIALOG SPOOL
Dialog steps 5 677 10
Response time total (s) avg (ms) 0 423 0 5 625 49
Summary for tasktype DIALOG Used work processes Dialog steps with DB calls waiting > 50 ms with roll in Avg elapsed time Avg not in workp. Avg response time CPU time wait time load time DB request time Roll time in out
4 677 372 58 674 2.233,8 1.608,4 625,4 141,2 46,7 78,3 56,8 37,6 41,1 ms ms ms ms ms ms ms s s
First record time Last record time Elapsed time Response time CPU time wait time oad time DB request time Roll in out
11:10:44 11:19:08 504,0 423,4 95,6 31,6 53,0 38,4 720 719 s s s s s s
Monitoring of the New Memory Management System

CCMS: Control Panel: Operation Mode/Server Status
Edit Select Monitoring Control Utilities System Help
Choose Refresh Choose OP mode Alert-Details Previous View Next View
Active op mode: DAY-OPERATION Memory management absolute value Sorted by Servername, active/inactive
Server-Name
Extended Memory (MB) Process-local Memory Amount of Total Peak Current in Current Peak restarted WP's used used process used used DIALOGnon DIA 319 63 319 169 63 179 60 13 98 7 10 13 0 KB 0 KB 4873 KB 0 KB 62 MB 30 MB 0 1 0 0 0 0
hs0301_P30_20 hs0313_P30_20 hs0315_P30_20
P30(1) hs0313 OVR 04:45PM SAP America, Inc. prf3.ppt December, 1996
Page 28
Memory management >= 3.0C

ztta/roll_first ztta/roll_extension ztta/roll_area ztta/roll_first abap/heap_area_dia
DIA
Roll
EM
EM
Roll
Heap
abap/heaplimit
em/initial_size_MB
Extended Memory
ztta/roll_area
abap/heap_area_nondia
BTC UPD
Roll
abap/heaplimit
Heap
EM
ztta/roll_extension
Configuration Strategy
What can be done to save main memory ? decrease size of roll- and paging buffers rdisp/ROLL_SHM smaller up to 0 rdisp/PG_SHM smaller up to 0 store more data in roll area ztta/roll_first bigger up to (ztta/roll_area - 1MB) in extreme main memory shortage, use traditional paging abap/use_paging = 1
Page 29
High performance configuration with big main memory and swap space reduce data in roll area ztta/roll_first smaller up to 1 (1 byte !!) provide more extended memory em/initial_size_MB bigger, about 7MB/user full buffering of roll- and paging file rdisp/ROLL_SHM bigger, about 2MB/user rdisp/PG_SHM bigger, about 1MB/user
What can be done to save swap space ? more efficient use of roll area ztta/roll_first bigger, ztta/roll_area - 100KB return process local space abap/heaplimit smaller, up to 1 MB no buffering of roll- and paging file rdisp/ROLL_SHM 0 rdisp/PG_SHM 0 reduce extended memory em/initial_size_MB smaller, see utilization in ST02 or RZ03
Page 30
What to do with huge data sizes in background jobs ? increase size of process local space abap/heap_area_nondia bigger, up to OS-limitations abap/heap_area_total bigger, enough swap space must available increase size of quota for extended memory ztta/roll_extension bigger There are platform dependent OS limitations on the size of a processes' address space and on the size of the process local data segment. Limitations on process data segment AIX: 256 MB SINIX: 384 MB HP-UX: about 640 MB SUN, NT: nearly 2000MB - (shared + extended) OSF1: > 4000 MB (64 Bit Architecture) In extreme situations, configure a dedicated background server with traditional paging abap/use_paging = 1
Workload Analysis
ABAP/4 Trace Tool - Runtime Analysis
Page 31
Runtime Analysis: Initial Screen SAP ABAP/4ABAP/4 Trace (ATRA) Tool Runtime analysis Edit Goto Utilities System Help ?
Tips & Tricks
Transaction Program
ZFPATRA6
Execute
Options
With subroutines
Restriction to
Function module With internal tables All With technical DB info
Runtime analysis ZFPATRA6 test program: before tuning. 17.04.96 16:10:23 File size : 255 kB
Other trace files ...
Analyze
Delete
TCC(1)
hsext01
OVR
11:38AM
Runtime Analysis Evaluation: Overview Runtime analysis Edit Goto Utilities System Help ? Hit list Tables Group hit list Hierarchy
Program User
ZFPATRA6 = ATRA test program: before tuning. MDLR Execution time in microseconds: ABAP/4 Database R/3 SYSTEM 0% 50% 100%
Time Date
16:10:23 17.04.96
311.683 = 52,6% 264.986 = 44,7% 15.648 = 2,6% 592.317 = 100,0%
Maximum memory requirement in kB Statistics entry Conversions Convert from type P Convert to type P Type P arithmetic Loaded programs
579 Number 210 104 104 469 4
TCC(1)
hsext01
OVR
11:38AM
Page 32
Analysis Evaluation: Overview SAP ABAP/4Runtime Trace (ATRA) Tool Runtime analysis Edit Goto Utilities System HelpFilter Display ? Hit list Tables
Display
Program User Applicat. program ZFPATRA6 = ATRA test program: before tuning. Time 16:10:23 System program MDLR Date 17.04.96 Database interface (Open SQL) Database interface (Native SQL) Execution time in microseconds: Database and buffer operations Matchcode processing ABAP/4 Screen system processing steps 311.683 = 52,6% Database 264.986 = 44,7% Memory and program runtime management R/3 SYSTEM 15.648 = 2,6% 0% 50% 100% 592.317 = 100,0%
Group hit list
Hierarchy
Net times are recalculated accordingly

Maximum memory requirement in kB Statistics entry Conversions Convert from type P Convert to type P Type P arithmetic Loaded programs 579 Number 210 104 104 469 4
TCC(1)
hsext01
OVR
11:38AM
Analysis Evaluation: Hierarchy SAP ABAP/4Runtime Trace (ATRA) Tool Runtime analysis Edit Goto Utilities System Help ?
Choose Absolute <-> % Scre text
Mem.Req.(kB)
292 243 253 253 257 257 257 272
Gross
514.230
Net
Program Lv
0 1 1 2 1 2 1 1
Call hierarchy
Submit report ZFPATRA6 Perform(Ext) BEFORE_EVENT Open Cursor TABNA DB Operation OpenCursor Fetch TABNA DB Operation FetchCursor Collect ITABNA Fetch TABNA
192.385 SAPMS38T 70 ZFPATRA6
897
896 ZFPATRA6 1 ZFPATRA6
67.537
151 ZFPATRA6 67.386 ZFPATRA6 90 ZFPATRA6 58 ZFPATRA6
TCC(1)
hsext01
OVR
11:38AM
Page 33
Analysis Evaluation: HitTool List SAP ABAP/4 Runtime Trace (ATRA) Runtime analysis Edit Goto Utilities System Help ?
Choose Absolute <-> % Gross <-> Net Scre text
Call
Submit report ZFPATRA6 Fetch TABNA
In progr
SAPM38T ZFPATRA6
No.
1 635 6 634 1 369 1
Gross
Net
Type
Display filter group
514.230 192.385 220.694 27.562 193.132 118.870 3.378 3.282 897 896 DB OpenS DB DB OpenS DB
DB Operation FetchCursor ZFPATRA6 Collect ITABNA Sort TABNASHOW Collect TABNASHOW Open Cursor TABNA ZFPATRA6 ZFPATRA6 ZFPATRA6 ZFPATRA6
TCC(1)
hsext01
OVR 11:38AM
SAP ABAP/4 Trace Tool

Runtime Analysis Evaluation: Hit List Runtime analysis Edit Goto Utilities System Help ?
Choose Absolute <-> %
Table
TABNA Total
#Access
637 637
Gross
221.596 221.596
Class
Buffering
SQL tab.
Description
TRANSP
BC: Training table
TCC(1)
hsext01
OVR 11:38AM
Page 34
SAP ABAP/4 Trace (ATRA) Tool

Call of Table accesses TABNA
No
Call
428 Open Cursor TABNA 99 412 261 503 Fetch TABNA Fetch TABNA Fetch TABNA Fetch TABNA
Gross
897 227 160 67.537 30.486
Net
896 227 160 151 125
Program
ZFPATRA6 ZFPATRA6 ZFPATRA6 ZFPATRA6 ZFPATRA6
Include
30 30 30 30 30
Line
ZFPATRA6 ZFPATRA6 ZFPATRA6 ZFPATRA6 ZFPATRA6
TCC(1)
hsext01
OVR
11:38AM
Runtime Analysis (ATRA) Evaluation: Source Code SAP ABAP/4 Trace Tool Runtime analysis Edit Goto Utilities System Help ?
Editor Program ZFPATRA6 Call Open Cursor TABNA
..... 00029 00030 00031 00032 00033 00034 00035 ..... 00041 00042 00043 ..... 00046 00047 00048 00049 00050
REFRESH ITABNA SELECT * INTO ITABNA FROM TABNA ORDER BY SALES POSTCD. APPEND ITABNA SORTED BY COUNTRY. ENDSELECT. * Countries WRITE: /(39) 'Umsatzzahlen der EG Lnder' COLOR 6 NO-GAP. SKIP. LOOP AT ITABNA. CHECK NOT ITABNA-COUNTRY CP 'USA' AND NOT ITABNA-COUNTRY CP 'CAN' AND MOVE-CORRESPONDING ITABNA TO TABNASHOW. APPEND TABNASHOW SORTED BY COUNTRY. ENDLOOP. * SORT TABNASHOW BY COUNTRY.
TCC(1)
hsext01
OVR
11:38AM
Page 35
Runtime of ABAP/4 Statements SAP ABAP/4 Trace (ATRA) Tool List Edit Goto System Help ?
Choose
SQL Interface Select ... Where vs. Select + Check Select single vs. Select-Endselect Select aggregates String manipulation Special operators in IF (CA, ...) String concatenation II Deleting leading spaces
Select with index support Select ... Into Table t Select-Endselect vs. Array-Select
String concatenation String split String length
Internal Tables
Building sorted tables Building tables without duplicates Different forms of key access Building condensed tables Linear vs. binary search Secondary indices
TCC(1)
hsext01
OVR
11:38AM
Runtime Analysis: Initial Screen SAP ABAP/4ABAP/4 Trace (ATRA) Tool Statements Edit Goto System Help ?
Previous example Measure runtime List examples Display data
Select + Check Statement SELECT * FROM VERI_CLNT. CHECK: VERI_CLNT-ARG1 = ' 7 ' ENDSELECT .
Select with Where condition SELECT * FROM VERI_CLNT. WHERE ARG1 = ' 7 ' ENDSELECT .
microsec
18.767
microsec
1.587
* Always specify your conditions in Where-clause instead of * checking them yourself with check-statements. * The database system can then use an index ( if possible ) and the * network load is considerably less.
TCC(1)
hsext01
OVR
11:38AM
Page 36
Questions ?
Page 37

SAP Architecture

Uploaded by

Document Informationclick to expand document informationBest document understand the basic SAP Architecture

Document Informationclick to expand document information

Copyright:

Available Formats

SAP Architecture

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

SAP Architecture

Uploaded by

Copyright:

Available Formats

R/3 Workload Analysis

SAP America, Inc. prf3.ppt December, 1996

R/3 Performance Monitoring Tools

Expensive SQL statements

Memory Management for R/3 3.x

ABAP/4 trace tools

SAP America, Inc. prf3.ppt December, 1996

R/3 Architecture Review

SAP America, Inc. prf3.ppt December, 1996

R/3 System Review - Dialog Processing

Roll buffer Page buffer Page file Roll file Network

SAP Shared Pool Buffers

Database process Database Buffers

SAP America, Inc. prf3.ppt December, 1996

R/3 System Review - Update Processing

Mr. Jack Knife

Mr. Jack Knife

The R/3 Performance Timing Statistics

Roll in DB Time Network

Dispatch Time Response Time

SAP America, Inc. prf3.ppt December, 1996

R/3 Performance Monitoring Tools

SAP America, Inc. prf3.ppt December, 1996

Main Monitors for Analysis

SAP Buffers Monitor

Process and OS Monitors

SAP America, Inc. prf3.ppt December, 1996

The Workload Monitor (st03)

SAP application server

Workload Summary of an Instance

Response times for all task types

4.950 885,1 ms 147,7 ms 5,3 ms 6,4 ms 357,0 ms 35,6 kB 6.284 4.784

SAP America, Inc. prf3.ppt December, 1996

The Workload Monitor (st03)

Transaction Profile Information

Program or Dialog Time Steps

SAP America, Inc. prf3.ppt December, 1996

The Workload Monitor (st03)

Top Time Profile

SAP America, Inc. prf3.ppt December, 1996

Process Overview Monitor (sm50)

Maria: Close these boxes off with a line

The OS Monitor (st06)

Count Load average

1 min 5 min 15 min Context switches/sec

1 1,78 1,40 1,31 2.927 700 0 0 3.063.532

The Database Monitor (st04)

162.024 95 98 100 60.589.413 586.102 4.153

20.480 42.457.752 2.771 0 23.091.509 2.375.072

The Database Monitor (st04)

Exclusive session-lock situations (no shadow process)

Host with Lock,PID

SAP America, Inc. prf3.ppt December, 1996

Database analysis Cho. 02.05.1996

Data Base Performance: Shared SQL Goto Monitor System Help

Shared Cursor Cache

( last reset at 02.05.1996 09:23:30 )