Test 234: AIX 5L Performance and Systems Tuning

Section 1 - Measurements/Baseline

• Determine successful operating levels for stable/comparable system
  • Run the oslevel command to determine current os version
  • Run the lslpp command to list installed software
  • Run errpt to determine if there are any unresolved hardware problems
• Take snapshot/baseline of system
  • Install perfagent.tools fileset
  • Install bos.perf fileset
  • Install bos.adt.samples fileset
  • Verify the latest filesets for all the above steps, if not, obtain latest and install
  • Identify time to take performance measurements
  • Run perfpmr to gather operating statistics
  • Run no, nfso, vmtune, schedtune to gather tuning parameters if appropriate
  • Run wlm (Workload Manager) to gather tuning parameters
  • Run topas (Performance Monitoring Tool) to gather tuning parameters
• Establish performance policies for new system (system to be tuned)
  • Determine which measurements are relevant
  • Define (via Service Level Agreements) what a "well behaved" system is

Section 2 - Performance Monitoring and Analysis

• Determine if CPU bound
  • Run vmstat, iostat, sar, or topas to gather cpu utilization information
  • Add %user and %sys to determine if cpu bound
• If cpu bound, run additional tools like tprof and ps to determine which process is causing the bottleneck Determine if memory bound
  • Run vmstat to gather memory utilization information
  • Look at pi and po output columns to determine if memory bound
  • Use /proc to determine if the system is memory bound
  • If memory bound, run tools like svmon and ps to determine which process is causing the bottleneck
• Determine if I/O bound
  • Run iostat or topas to gather disk I/O utilization information
  • Look at %iowait, %tm_act to determine if I/O bound
  • If I/O bound, run tools like filemon, fileplace and lslv to determine which process, disk, or adapter is causing the bottleneck
• Determine if network bound
  • Run netpmon, netstat and/or nfsstat to gather network utilization information
  • Run entstat or fddistat to gather adapter-specific information
  • Look at output to determine if network bound
  • If network bound, run tools like no, netpmon, nfso, lsattr to assist in determining the cause of the bottleneck

Section 3 - Tuning

• Tune application
  • Analyze application with profiling tools like tprof, gprof, and prof to identify potential code optimization opportunities
• Tune CPU
  • Balance system workload
  • Tune scheduler using nice/renice to give different priorities to running processes (prevent cpu hogging)
  • Tune scheduler algorithm using schedtune to fine tune priority formulas, setting timeslices
• Tune memory
  • Add more real memory to increase available memory
  • Tune virtual memory manager using vmtune command
  • Tune memory overcommitment algorithm using schedtune command to prevent thrashing
• Tune I/O
  • Place data on physical disks appropriately (striping) using tools such as reorgvg, migratepv, chlv
  • Use vmtune to modify file system, lvm and paging parameters
  • Use chgsys to tune high/low watermarks for pending write I/O's (I/O pacing)
  • Place JFS logs on different physical disks from filesystem data
  • Defragment files, make them contiguous
• Tune network
  • Tune transmit and receive queues using chdev
  • Use no to tune TCP/IP options
  • Use ifconfig or no to adjust MTU size appropriate for your network topology
  • Use /etc/netsvc.conf or NSORDER environment variable to override default order for name resolution
  • Tune NFS options (if appropriate)

Section 4 - System Sizing

• Run performance monitoring tools to gather additional information
• Analyze data to predict future system requirements
• Use sizing tools
  • Analyze results of sizing tool

 

 

1.       Which of the following commands should be used to obtain information about the level of the operating system and maintenance level?

A.      lslpp

B.      uname

C.      oslevel

D.      rpcinfo

2.       An AIX system must be evaluated and tuned. In order to use a set of performance tools (e.g. filemon, svmon, and tprof) which of the following packages should exist on the system in question?

A.      Java.adt

B.      bos.perf

C.      bos.adt.utils

D.      perfagent.tools

3.       When paging, cache, and Translation Lookaside Buffer (TLB) metrics or statistics are analyzed for performance, which of the following resources is being analyzed?

A.      CPU

B.      Memory

C.      Disk I/O

D.      Network I/O

4.       When monitoring a system with vmstat, which of the following numbers under the cpu heading indicate that the system is probably CPU bound?

A.      The value of wa greater than 70

B.      The value of id is greater than 70

C.      The sum of us and sy consistently between 99 and 100

D.      The sum of id and wa consistently between 99 and 100

5.       Which of the following commands should be used to show the number of system calls per second that are being executed?

A.      pstat

B.      vmstat

C.      filemon

D.      lockstat

6.       Refer to the exhibit below. While a user is compiling a C program, "vmstat 120 10" is run to determine the cause of a performance problem. Given the vmstat output as shown in the exhibit indicates an I/O bottleneck, which of the following commands should be run next to get more information about the problem?

A.      lsps

B.      tprof

C.      iostat

D.      vmtune

7.       Which of the following commands is most useful in collecting data to determine a network performance problem?

A.      iostat

B.      lpstat

C.      netstat

D.      vmstat

8.       Which of the following commands should be used to identify packet sequence problems?

A.      tprof and gprof

B.      netstat and iostat

C.      lsattr and ifconfig

D.      iptrace and tcpdump

9.       Which of the following general operation techniques will best provide a measurable performance improvement of a system?

A.      Attach SCSI devices in ID order.

B.      Perform backups at regular intervals.

C.      Perform regular preventative maintenance of devices with moving parts.

D.      Reschedule the processing to balance the workload.

10.   Which of the following is TRUE of the fileplace command?

A.      A file can be placed in a specific location.

B.      The fragment distribution on a specified file is shown.

C.      Fragmentation is removed by rearranging file fragments.

D.      The distribution of all the files on a specified disk are shown.

11.   Refer to the exhibits below. On an NFS server, "netstat -m" is run once and then again 15 minutes later. An extract of the output for both times is shown in the exhibits. Using the netstat outputs, which of the following conclusions is most appropriate to draw?

A.      The number of biod(s) processes should be increased.

B.      The number of nfsd(s) processes should be increased.

C.      The machine needs more physical memory.

D.      The machine needs more memory allocated for mbufs.

Exhibit 1.1.b.1 "1st netstat output"

dimensions 5.5in 4.5in

# date

Mon Jun 12 20:05:54 CDT 1995

# netstat -m

1385 mbufs in use:

96 mbuf cluster pages in use

2170 Kbytes allocated to mbufs

841 requests for mbufs denied

0 calls to protocol drain routines

Kernel malloc statistics:

Exhibit 1.1.b.1.x2 "2nd netstat output"

dimensions 5.5in 4.5in

# date

Mon Jun 12 20:20:04 CDT 1995

# netstat -m

1389 mbufs in use:

97 mbuf cluster pages in use

2180 Kbytes allocated to mbufs

982 requests for mbufs denied

0 calls to protocol drain routines

Kernel malloc statistics:

12.   Which of the following tools should be used to begin gathering time-stamp data on the system for a performance analysis?

A.      sar

B.      iostat

C.      netstat

D.      vmstat

13.   The backups for a system are being taken over the network to a TSM server. Unfortunately, they are taking too long and the service level agreement is not being met with the customer. Based upon the current configuration, which of the following tuning procedures should be performed to improve the file system's backup performance?

A.      Increase the maxfree value using vmtune.

B.      Increase the mult value using schedtune.

C.      Increase the queue length on the disk drives using chdev.

D.      Increase the tcp_sendspace/tcp_recvspace on your system and the TSM server.

14.   After a migration from an older Power server to a PowerPC server, users complain about decreased performance. Which of the following tools should be used to investigate the performance problem?

A.      netstat

B.      iostat

C.      perfpmr

D.      emstat

15.   An application is causing a response time issue. Which of the following can be used to determine the problem?

A.      filemon

B.      fileplace

C.      rmss

D.      netstat

Answer Key:

1.       C

2.       D

3.       B

4.       C

5.       B

6.       C

7.       C

8.       D

9.       D

10.   B

11.   D

12.   A

13.   D

14.   D

15.   A