Hitachi

uCosminexus Application Server System Design Guide

5.2.3 Estimating the resources used by performance tracer

This subsection describes the estimation of resources used by performance tracer for each OS.

Organization of this subsection

(1) In Windows

This section describes the estimation of resources used by performance tracer when using Windows.

(a) File size for shared memory

Calculate the file size (Unit: Bytes) for shared memory used by performance tracer for each PRF daemon. The formula for calculating is as follows:

  • Formula for calculating shared memory for each PRF daemon

    File-size-for-shared-memory = -PrfTraceBufferSize-specified-value × 1,024+18,496

(b) Disk occupancy of %PRFSPOOL%

Formula for calculating disk occupancy of %PRFSPOOL% is as follows:

  • Formula for calculating disk occupancy of %PRFSPOOL%

    Disk-occupancy = 2.0 MB

    + {(-PrfTraceBufferSize-specified-value + 20KB) × 5

    + -PrfTraceFileSize-specified-value × -PrfTraceCount-specified-value × r} × n

    + 224 KB × (256+m)

    + 224 KB × (64 + p)

    Legend:

    • n: Number of PRF daemons

    • m: Number of running performance trace output processes and number of performance trace output processes that did not terminate normally

      In a performance tracer, the internal trace is output to a file as the maintenance information for each performance trace output process. This file is created when a process starts, but the file remains if the process has abnormal termination. The file is deleted when a PRF daemon is invoked and every 24 hours after the PRF daemon is invoked, but 256 files remain without being deleted. Therefore, the maximum number of files is "256 + number-of-performance-trace-output-processes-executed-in-24-hours".

    • p: Number of commands and daemon processes used in the running performance analysis trace

      In a performance tracer, the internal trace is output to a file as the maintenance information for each command and daemon used in the performance analysis trace. This file is created when a process starts. The file is deleted when a PRF daemon is invoked and every 24 hours after the PRF daemon is invoked, but 64 files remain without being deleted. Therefore, the maximum number of files is "64 + number-of-command-and-daemon-processes-used-in-performance-analysis-trace-executed-in-24-hours".

    • r: Backup coefficient

      This value calculates the PRF trace backup. If you specify -PrfNoBackUp 0 in the invocation option for the cprfstart command, the backup coefficient is 2, and in all other cases, the backup coefficient is 1.

The above mentioned disk capacity is a rough indicator. As a result, create the %PRFSPOOL% with enough excess space.

To Page Top

(2) In AIX

You must consider the following values and set up the kernel parameter to use the performance tracer when using AIX. If the settings are not done properly, the processes of performance tracer cannot be invoked or while running the processes they might terminate abnormally due to insufficient resources. For the kernel parameter settings, see the manual of the OS being used.

(a) File size for shared memory

Calculate the file size for the shared memory (Unit: Bytes) used by performance tracer for each PRF daemon. The formula for calculating is as follows:

  • Formula for calculating shared memory for each PRF daemon

    File-size-for-shared-memory = -PrfTraceBufferSize-specified-value × 1,024+18,496

Set up the file size for shared memory using the environment variable EXTSHM. Set up in such a way so that shared memory allocated is greater than the value calculated in the formula for calculations.

(b) Number of file descriptors

Set up the number of file descriptors in nofiles of the /etc/security/limits file. Set up the number of file descriptors used when invoking the PRF daemon as 32 and above.

(c) Disk occupancy of $PRFSPOOL

The formula for calculating the disk occupancy of $PRFSPOOL is as follows:

  • Formula for calculating disk occupancy of $PRFSPOOL

    Disk-occupancy = 2.0 MB

    + {(-PrfTraceBufferSize-specified-value + 20KB) × 5

    + -PrfTraceFileSize-specified-value × -PrfTraceCount-specified-value × r} × n

    + 224 KB × (256+m)

    + 224 KB × (64 + p)

    Legend:

    • n: Number of PRF daemons

    • m: Number of running performance trace output processes and number of performance trace output processes that did not terminate normally

      In a performance tracer, the internal trace is output to a file as the maintenance information for each performance trace output process. This file is created when a process starts, but the file remains if the process has abnormal termination. The file is deleted when a PRF daemon is invoked and every 24 hours after the PRF daemon is invoked, but 256 files remain without being deleted. Therefore, the maximum number of files is "256 + number-of-performance-trace-output-processes-executed-in-24-hours".

    • p: Number of commands and daemon processes used in the running performance analysis trace

      In a performance tracer, the internal trace is output to a file as the maintenance information for each command and daemon used in the performance analysis trace. This file is created when a process starts. The file is deleted when the PRF daemon is invoked and every 24 hours after the PRF daemon is invoked, but 64 files remain without being deleted. Therefore, the maximum number of files is "64 + number-of-command-and-daemon-processes-used-in-performance-analysis-trace-executed-in-24-hours".

    • r: Backup coefficient

      This value calculates the PRF trace backup. If you specify -PrfNoBackUp 0 in the invocation option for the cprfstart command, the backup coefficient is 2, and in all other cases, the backup coefficient is 1.

The above mentioned disk capacity is a rough indicator. As a result, create the $PRFSPOOL with enough excess space.

To Page Top

(3) In Linux

You must consider the following values and set up the kernel parameter to use the performance tracer when using Linux. If the settings are not done properly, the processes of performance tracer cannot be invoked or while running the processes they might as well terminate abnormally due to insufficient resources. For the kernel parameter settings, see the manual of the OS being used.

(a) File size for shared memory

Calculate the file size for shared memory (Unit: Bytes) used by performance tracer for each PRF daemon. The formula for calculating is as follows:

  • Formula for calculating shared memory for each PRF daemon

    File-size-for-shared-memory = -PrfTraceBufferSize-specified-value × 1,024+18,496

Set up the file size for shared memory in kernel.shmmax of /etc/sysctl.conf file. Set up in such a way so that shared memory allocated is greater than the value calculated in formula for calculations.

(b) Number of file descriptors

Set up the number of file descriptors in nofiles of /etc/security/limits.conf file. Set up the number of file descriptors used when invoking the PRF daemon as 32 and above.

(c) Disk occupancy of $PRFSPOOL

The formula for calculating the disk occupancy of $PRFSPOOL is same as AIX. Reference the formula for calculations of AIX.

To Page Top

Trademarks

All Rights Reserved. Copyright (C) 2022, Hitachi, Ltd.