DB-7: OpenEdge® Database Performance Tuning Gus Björklund Wizard, Vice President Technology.

DB-7: OpenEdge® Database Performance Tuning

Gus BjörklundWizard, Vice President Technology

© 2006 Progress Software Corporation2DB-7: OpenEdge Database Performance Tuning

Topics

Easy Stuff: The Basics Disk Layout Esoterica


First Things First

>> probkup foo>


“Out of the Box” Performance

>> proserve foo>


“Out of the box” Performance

30 00

100200300400500600700800

no tuning simple tuning

thro

ugh

put

(tps)


“Nine Easy Pieces” for Better Results


1: Buffer Pool Size

>> proserve foo -B 32000>


2: Spinlock retry count

>> proserve foo -B 32000 -spin 50000>


3: Start BI Log Writer (BIW)

>> proserve foo -B 32000 -spin 50000> probiw foo>


4: Start Async Page Writer (APW)

>> proserve foo -B 32000 -spin 50000> probiw foo> proapw foo> proapw foo>


5: Increase BI Log Block Size

>> proutil foo -C truncate bi \> -biblocksize 8> proserve foo -B 32000 -spin 50000> probiw foo> proapw foo> proapw foo>


6: Increase BI Log Cluster Size

>> proutil foo -C truncate bi \> -biblocksize 8 -bi 4096> proserve foo -B 32000 -spin 50000> probiw foo> proapw foo> proapw foo>


7: Add BI Log buffers

>> proutil foo -C truncate bi \> -biblocksize 8 -bi 4096> proserve foo -B 32000 -spin 50000 \> -bibufs 25> probiw foo> proapw foo> proapw foo>


8: Initial Database Layout

d "Schema Area" /users/gus/atm.d1d "atm":7,64;512 /users/gus/atm_7.d1 f 2000000d "atm":7,64;512 /users/gus/atm_7.d2 f 2000000d "atm":7,64;512 /users/gus/atm_7.d3 f 2000000d "atm":7,64;512 /users/gus/atm_7.d4 f 2000000d "atm":7,64;512 /users/gus/atm_7.d5 f 2000000d "atm":7,64;512 /users/gus/atm_7.d6 f 2000000d "atm":7,64;512 /users/gus/atm_7.d7b /users/gus/atm.b1


8: Move Data Extents to Striped Array

d "Schema Area" /users/gus/atm.d1d "atm":7,64;512 /array/atm_7.d1 f 2000000d "atm":7,64;512 /array/atm_7.d2 f 2000000d "atm":7,64;512 /array/atm_7.d3 f 2000000d "atm":7,64;512 /array/atm_7.d4 f 2000000d "atm":7,64;512 /array/atm_7.d5 f 2000000d "atm":7,64;512 /array/atm_7.d6 f 2000000d "atm":7,64;512 /array/atm_7.d7b /users/gus/atm.b1


9: BI Log On Separate Disk

d "Schema Area" /users/gus/atm.d1d "atm":7,64;512 /array/atm_7.d1 f 2000000d "atm":7,64;512 /array/atm_7.d2 f 2000000d "atm":7,64;512 /array/atm_7.d3 f 2000000d "atm":7,64;512 /array/atm_7.d4 f 2000000d "atm":7,64;512 /array/atm_7.d5 f 2000000d "atm":7,64;512 /array/atm_7.d6 f 2000000d "atm":7,64;512 /array/atm_7.d7b /bidisk/atm.b1


Can you predict the results ?


Now Our Results Are

30

581

0100200300400500600700800

no tuning simple tuning

thro

ugh

put

(tps)


Effect of Tuning -spin

133 120

436

680 679

0100200300400500600700800

spin setting

thro

ugh

put

(tps)


Effect of Tuning -B

260295

348 332

050

100150200250300350400450500

32,000 64,000 128,000 256,000

-B setting

thro

ughp

ut (t

ps)


Disk Layout


About disks

Inexpensive !! One drive can do only one thing at a time

• Many smaller drives better than one large

15,000 rpm drives SCSI still better than SATA II Desktop drives rated 8 hours on, 16 off Avoid cheesy controllers Spares


Multiple Disk Drives


Stripe and Mirror


Put BI Log on own disk


Use Type II Data Areas


Type I vs Type II Areas

100

140

020406080

100120140160180200

Type I DataArea

Type II DataArea

Relative Performance

thro

ugh

put

(perc

ent)


Esoterica


Networking

Raise message size (-Mm) to 16k Use Gigabit Ethernet Configure Ethernet for FDX Multiple brokers

• One for 4GL, one for SQL

• Others

Raise max servers (-Mn) and (-Mpb) Lower clients per server (-Ma)


Temporary files (-T)

Local drives, not on network Spread users over multiple disks


Use Shared Procedure Libraries

One copy of r-code for many users Use saved memory for other things

• like higher -B

All sharers have to be on same machine• Don’t need on client machines


Creating Records

Template holds default column values Creating a record makes a copy of template

• CREATE cust

Fill in all column values immediately• Use one ASSIGN

• Avoids an update later

• Avoids fragmentation if default record is very small


Linux File Systems

292313

267

348

0

50

100

150

200

250

300

350

400

ext3 JFS XFS ReiserFS

thro

ug

hp

ut

(tp

s)

ReiserFS results wildly erratic - don’t use !


Other filesystems

Operating system File system

Solaris VxFS or UFS

AIX JFS or JFS2

HP-UX NOT HFS

Windoze NTFS

Linux ext3, XFS, JFS


Windows Tips (part 1)

Remove crapware• use Ad-Aware or Spybot S&D

• use Firefox, NOT IE

Turn off• services you don’t need

– http://majorgeeks.com/page.php?id=12

• startup items you don’t need

Black screensaver only Get rid of Office (at least indexer)


Windows Tips (part 2)

NTFS filesystem• Never FAT-32 or FAT-16

Increase NTFS cluster size• format <drive>:/fs:ntfs /A:8k

Turn off• short filenames

• file sharing

• virus scanner !

• all GUI decorations

“Maximize Throughput for Network Applications”


SQL Tips

Add more indexes Run UPDATE STATISTICS Use ODBC and JDBC API’s well

• See DataDirect.com/developerfor technical papers

Upgrade to 10.1


More Esoterica

Buy more memory - cheapest way to improve Use 64-bit OpenEdge Mount filesystems with noatime Lock shared memory (-pinshm) Use “deadline” i/o scheduler on Linux We like 3ware controllers for mid-size boxen Set shmmax to 134,217,728


Summary

Getting good database performance is easy


More info

My tuning guide on Conference CD PEG (www.peg.com) Dan Foreman’s guides (info at expo) Related sessions:

• COMP–12: Load/Performance Testing Tools and Techniques

• DB–2: Tuning OpenEdge SQL: Boosting Your SQL Application Performance

• DB–5: OpenEdge Database Performance Made Easy• DB–8: Highly Parallel Dump and Load• DB–9: Pick an Index, Any Index...


WantAnswers


DB-7: OpenEdge® Database Performance Tuning Gus Björklund Wizard, Vice President Technology.

Documents