AIX System Administration III - Performance Management Course

This course is designed for System Administrators. It provides delegates with the skills to monitor, analyse and tune AIX subsystems for optimal performance including concepts, strategy and implementation of the AIX Workload Manager.

Delegates should be experienced System Administrators and should have previously attended AU14GB (AIX System Administration I: Implementation) and AU16GB (AIX System Administration II: Problem Determination).

Topics covered on the 5 day AIX System Administration III - Performance Management course

Performance Overview

• Introduction
• System performance
• Performance tuning process
• Resource allocation
• System architecture
• Program execution hierarchy
• Components of system performance
• Processor performance
• Memory hierarchy
• System dynamics/attributes
• Tools for performance evaluation
• Performance metrics
• Standard UNIX and AIX performance tools
• AIX performance toolbox
• 32/64-bit performance and architecture
• 32/64-bit process address space
• PCI bus layout
• CPU internal structure

Tuning CPU Usage

• Processes and threads
• The life of a process
• Scheduling policies
• Round robin
• Process and thread policy
• Scheduling formulas
• Using schedtune
• Scheduler run queue
• CPU timeslice
• Mode switching
• The iostat, sar, ps, acctcom, tprof and pstat commands
• Instruction emulation
• The emstat command
• Speeding up user name/user ID resolution
• Administration tuning techniques

Virtual Memory Performance

• VMM technology
• More segment types
• More virtual memory concepts
• Computational vs file memory
• Page stealer
• VMM memory load control
• The vmstat, ps and svmon commands
• rmss-an advanced memory simulator
• Use rmss as driver
• Interpreting rmss results
• System tuning techniques
• The vmtune command
• Tuning VMM page replacement
• Tuning file memory usage
• schedtune
• Is the system thrashing?
• Paging space
• Tuning paging space thresholds
• Early versus late page space allocation
• Deferred page space allocation
• Using shared memory
• Extended shared memory
• Memory leaks

Physical and Logical Volume Performance

• Storage concepts
• Adapter technology
• RAID storage
• The logical volume manager
• Logical volume striping
• Logical volume attributes that affect performance
• Monitoring PV and LV usage
• Using iostat and filemon
• Logical volume fragmentation
• Tuning PV and LV
• Disk/SCSI performance issues
• SCSI adapter configuration
• PV configuration/considerations
• Volume group recommendations
• Logical volume organisation for highest LVM performance
• Tuning striped logical volumes for highest LVM availability
• Log logical volumes
• Tuning LVM using vmtune

File System Performance

• File system I/O layer
• Fragmentation and performance
• Compression considerations
• Using filemon and fileplace
• File fragmentation and disk placement
• Tuning applications for I/O
• Asynchronous I/O
• fsync and sync
• Direct I/O
• JFS log
• Using raw disk I/O
• Tuning file systems I/O performances
• Reorganising a file system
• Defragmenting an file system
• Tuning sequential read ahead - minpgahead and maxpgahead
• VMM write-behind
• Tuning file syncs
• I/O pacing
• Miscellaneous parameters

Network Performance

• Network interfaces
• Internet protocol architecture
• Mbufs and clusters
• Ports and sockets
• Sockets (TCP/UDP)
• Maximum transmission unit
• TCP maximum segment size
• Send/receive queues
• NFS daemons
• Network performance issues
• The netstat and netpmon commands
• TCP and UDP tuning
• Tuning MTU and queues
• sb-max, tcp/udp_sendspace, tcp/udp_recvspace
• Using NSORDER and /etc/netsvc.conf in name resolution
• NFS tuning concepts
• The nfsstat command
• Tuning the number of nfsds
• /etc/filesystems and NFS
• The nfso command
• Tuning disks and hardware
• CacheFS
• Configuring CacheFS

Additional Tools and Services

• Trace
• The trcrpt command
• Using trace to detect trashing and other resource constraints
• PDT components
• Customising PDT
• Collection, retention and reporting
• Severity level and notification
• Thresholds
• Error reporting
• The perfpmr command
• PTX6000 and xmperf
• Utilities
• The topas command

Performance Case Study

• What is a performance problem?
• Problem description
• Collecting performance data
• Tuning strategy and flowchart
• Summary of tools
• Interpreting real data
• Reporting your findings
• Performance tuning summary

Symmetric Multiprocessing

• Multiprocessing
• Asymmetric versus symmetric
• SMP definitions
• SMP characteristics
• SMP cache coherency
• SMP software characteristics
• SMP synchronization issue
• SMP synchronization (threads/interrupts)
• Lock types
• AIX V4 SMP locks
• Lock penalties and granularity
• MP safe versus MP efficient
• Processor affinity issues/values
• SMP scaling
• Parallelising an application
• SMP benefits, limitations and tools
• Binding a process
• Performance tools on SMP
• Lock statistics
• The lockstat command
• System locks
• Lock contention
• Lock tuning using schedtune

Workload Manager Overview

• Server consolidation
• Application interaction
• System/physical partitioning
• Logical partitioning (LPAR)
• AIX Workload Manager
• LPAR and WLM
• Workload management
• AIX WLM functionality
• WLM: Classes, tiers, rules, shares and limits

Workload Manager Administration

• WLM property files; classes, limits, shares and rules
• WLM configuration
• Adding/deleting/listing a class
• Create/delete/list rules
• WLM modes of operation
• Start/stop/update WLM
• Checking the configuration with wlmcheck
• Monitoring WLM using wlmstat, svmon, ps and topas
• Passive and active mode
• Multiple WLM configurations