1 of 25

Power System Performance�&�Capacity Planning�

All trademarks and registered trademarks are the property of their respective owners.

2 of 25

Introduction / highlights

7 years – worked with mainframe MSP (EDS)
9 years – SE / Disk product manager – STC
4 years – mainframe performance consultant
28 years – Started and ran Midrange Performance Group (MPG)

Developed and sold to IBM PM400 – 1993
Developed and sold Performance Navigator - 2000 V1
Licensed Performance Navigator to IBM Techline for Capacity Planning – 2004
Acquired by Helpsystems in 2018

4 and counting – IBM Power performance consultant

2

3 of 25

Agenda

Best Practice Methodologies
Data Collection

NO DATA NO ANALYSIS

Managing historical performance data
Viewing / Analyzing historical performance data
Problem Determination
Management Reporting
Capacity Planning
Q & A

3

4 of 25

Power Performance & Capacity Planning

Best Practice Methodologies

Collect historical performance data

On ALL partitions (IBMi, VIOS, AIX or Linux)

It’s about managing a frame (serial#) not just a partition

Consistent collection intervals / cycling

IBMi 15-minute default
Best practice is 5-minutes
AIX / Linux – nmon 5-minutes cycle at midnight for 24 hours (288 intervals)

In a perfect world, 1 year of data
Infrastructure (CPU, disk, memory, PowerVM, etc)
Application resource consumption

Answers the who / what / when causes the performance issues

4

5 of 25

Power Performance & Capacity Planning

Best Practice Methodologies

Clearly document any performance issue or Capacity Planning goals

Performance issue

Slow Response time
batch job run time
batch jobs throughput
Adding new applications
Adding new partitions
etc

5

6 of 25

Power Performance & Capacity Planning

Best Practice Methodologies

Clearly document any performance issue or Capacity Planning goals

Capacity Planning goals

Solving performance issue
End of lease
OS upgrade
Server consolidation
Moving to cloud (HA/DR, Prod)
Mergers, acquisition, adding new application, etc
Saving $$$
Losing Data Center
etc.

6

7 of 25

Power Performance & Capacity Planning

Best Practice Methodologies

Clearly document any performance issue or Capacity Planning goals

Capacity Planning - questions

Deciding which workload data to use?

Peak day (24 hours) in past ‘x’ days
Average day over ‘x’ days

Deciding how much Room for Growth you want?

By partition?
By frame?
By Virtual Shared Pool?

Scale out or Scale up?
Disk strategy

Internal or external?
Encryption / compression?
HA/DR/Flash Copy strategy?

7

Answers will be different depending on:

On-prem
Prod vs HA/DR
On-Prem Dynamic Capacity (PEP2)
Migrating to Cloud

Different for each Cloud vendor

Software Licensing

8 of 25

Power Performance & Capacity Planning

Data Collection

IBMi Collection Services

Defaults

*MGTCOL object 120 hours – 5 days
QAPM…. Files 10 days
Historical summary – 12 days 1-hour averages (off by default until V7R5)

.25, .5, 1, 5, 15, 30, 60 options

Historical details – 7 days, top 20 jobs

*all or ‘xx’ number of jobs

8

9 of 25

Power Performance & Capacity Planning

Data Collection

AIX / VIOS

Defaults

Topas data – 7 days

Better option

NMON – add to crontab to cycle at midnight with 5 minutes intervals (snaps)
Nmon chart / topas chart – open source – generates HTML for each day

Linux

NMON – open source

HMC

Enable Performance Collection in advance section of profile

Collects CPU uncapped and Virtual share pool data

9

10 of 25

Power Performance & Capacity Planning

Managing historical performance data

IBMi

IBMi Collection Services configuration
Regular save files of QPFRDATA

Library, QAPM* files or *MGTCOL object

Open Source

Lpar2rrd, Nagios, Zabbix, etc

Saas

Galileo, Maxava Mi8, etc

Solutions

Performance Navigator – Historical host code is FREE
Robot Monitor
etc

10

11 of 25

Power Performance & Capacity Planning

Managing historical performance data

AIX

Topas data

Be default keeps 7 days
Can be increase via the topasrec –r flag

Nmon data

Customer managed

Open Source

Njmon, Lpar2rrd, Nagios, Zabbix, etc

Saas

Galileo, etc

Solutions

Performance Navigator
Robot Monitor

11

12 of 25

Viewing / Analyzing historical performance data

IBMi

IBM Performance Tools (LLP)

Green screen interface to data – one member at a time
Go perform menu

Navigator for i

GUI interface to one member at a time
GUI interface to summarized historical data
GUI interface to SQL services commands

iDoctor – job watcher, disk watcher, PEX – licensed product
SaaS – web interface to data in cloud
Open Source – web interface to data on-prem
Solutions – Windows or web interface

12

13 of 25

Viewing / Analyzing historical performance data

AIX / Linux

Topas commands
Nmon commands
Open source – web interface to data on-prem
Saas – web interface to data in cloud
Solutions – Windows / web interface

13

14 of 25

Viewing / Analyzing historical performance data

AIX / Linux

Topas commands
Nmon commands
Open source – web interface to data on-prem
Saas – web interface to data in cloud
Solutions – Windows / web interface

14

15 of 25

Viewing / Analyzing historical performance data

3 main purposes for viewing/analyzing data

Problem determination
Management reporting
Capacity Planning

15

16 of 25

Problem Determination

17 of 25

View / Analyzing Historical performance data

17

Problem Determination

Clearly define the issue
Review historical data to determine significant changes
Determine if hardware configuration is part of the issue
If hardware is part of the issue, the determine options
If hardware is performing as configured, then

Determine which application are consuming the resources
Determine if any tuning can help
Provide application team with resources consumed

Demo using Performance Navigator

18 of 25

Management Reporting

19 of 25

View / Analyzing Historical performance data

19

Management Reporting

Produce reports for management on regular bases

Daily, weekly, monthly

Shows current data plus historical trending
Metrics compared to best practice guidelines
Hopefully automatically produced
By, partition, by frame, by Enterprise

Demo using Performance Navigator

20 of 25

Short Break – 10 minutes�

21 of 25

Capacity Planning for IBM i AIX/Linux

22 of 25

Capacity Planning for Power

22

Normally done as a CPaaS (Capacity Planning as a Service)

Requires skills in hardware, OS, storage, PowerVM, cloud architecture + pricing options (hardware and software)
Requires IBM, BP or cloud vendor to run configs and get pricing
Multiple options using same data:

Update existing Power hardware (add cores, memory, disk, etc.)
Migrate to new on-prem hardware – (Power 8 to Power 9 or Power 20)
Upgrade to Power 9 or 10 and use Dynamic capacity (PEP2) – on-prem cloud like pricing
Migrate to cloud – every cloud is different

23 of 25

Capacity Planning for Power

23

Ultimate Goal

Recommend best cost/performance Power model(s)
Recommend number of active cores and/or VSPs

Based on given Room for Growth
PEP2 – base license + recommended credits for break even over ‘x’ years

Recommend PowerVM settings (desired and virtual setting)
Recommend installed / active memory
Recommend assigned memory for each partition
Recommend disk configuration – space / performance

Internal disk – SSDs, NVMe, etc
Internal disk – Raid 5, 6, mirrored
External disk – Provide input to StorM for verification

Demo using Performance Navigator

24 of 25

Power Performance Analysis / Capacity Planning

24

Q & A

25 of 25

Thank you for joining us

All trademarks and registered trademarks are the property of their respective owners.