caccelstat Command

Purpose

Reports statistics related to coherent accelerators for the entire system, or for each accelerator and process.

Syntax

System-wide aggregate statistics

caccelstat -a [-t count] [-i interval]

Context statistics

caccelstat -c [-p pid -d device] [-t count] [-i interval] [-@ wparname]

Aggregate device or Accelerator Function Unit (AFU) statistics

caccelstat -d [device] [-t count] [-i interval] [-@ wparname]

Aggregate process context statistics

caccelstat -p [pid] [-t count] [-i interval] [-@ wparname]

Aggregate device kernel context statistics

caccelstat -k [device] [-t count] [-i interval] [-@ wparname]

Description

The caccelstat command is used to monitor coherent accelerator activity in the system. It can report aggregate statistics for all Coherent Accelerator Processor Interface (CAPI)-enabled processes, or it can show statistics for each CAPI-enabled process. It can also show statistics for each Accelerator Function Unit (AFU), where AFU is a coherent accelerator device such as /dev/caccel0.

System-wide aggregate statistics

The system-wide aggregate statistics report contains the following information:

Number of CAPI-enabled processes
Number of accelerators in the system
Number of user contexts
Number of kernel contexts
Number of master contexts
Number of page faults
Number of segment table (STAB) faults
Number of AFU errors
Number of AFU exceptions
Number of AFU signals
Number of AFU stalls

Context statistics

This report contains context statistics, one line per context. If the process ID (pid) and the device are not specified, all registered user contexts in the system are reported. Kernel contexts are omitted. If either the pid or the device is specified, contexts for that pid or device is reported. If both pid and device are specified, only the contexts pertaining to selected pid and device are reported.

Report contains following information:

State

Context state.

This field can have the following values:

S: The context is in suspended state.
R: The context is in running state.
D: The context is in detaching state.
A: The context is in attaching state.
E: The context is in error state.

For a master process, M is combined with the state. For example, if the context of the master process is in the running state, the state field displays MR.

Note: The running state (R) does not imply that the AFU is running this context. It means that it is not in any of the other states.

pid

Process ID. This column is skipped if pid is specified in the option.

pf

Number of page faults.

spf

Number of STAB faults.

pi

Number of page-ins.

err

Number of process errors.

exc

Number of process exceptions.

sig

Number of process signals.

stalls

Number of AFU stalls due to exception.

aur

Accelerator Utilization Register (AUR) value of this process. If AUR is not supported, this field shows 0.

device

Device name. This column is skipped if device is specified in the option.

Aggregate device (AFU) statistics

This report contains aggregate statistics for an AFU across all process contexts. If the device is not specified, statistics for all AFUs in the system is reported. This report includes following columns, one line per device:

nctx: Number of registered process contexts.
pf: Number of page faults.
spf: Number of STAB faults.
pi: Number of page-ins.
err: Number of AFU errors.
exc: Number of AFU exceptions.
sig: Number of AFU signals.
stalls: Number of AFU stalls due to exception queues being full.
device: Device name. This column is skipped if the device is specified in the option.

Aggregate process context statistics

This report contains aggregate and per process context statistics for a CAPI-enabled process. The pid is the AIX process ID. If the pid is not specified, statistics for all the CAPI-enabled processes in the system is reported.

This report includes following column, one per context:

pid: AIX Process ID. This column is skipped if the device is specified in the option.
nctx: Number of registered process contexts.
pf: Number of page faults.
spf: Number of STAB faults.
pi: Number of page-ins.
err: Number of AFU errors.
exc: Number of AFU exceptions.
sig: Number of AFU signals.
stalls: Number of AFU stalls due to exception queues being full.
aur: Accelerator Utilization Register value for this process. If AUR is not supported, this field shows 0.

Aggregate device kernel context statistics

This report contains aggregate or per device kernel context statistics. If no device is specified, statistics for all AFUs are reported. This report includes following column, one per device:

pf: Number of page faults.
spf: Number of STAB faults.
pi: Number of page-ins.
err: Number of AFU errors.
exc: Number of AFU exceptions.
device: Device name. This column is skipped if the device is specified in the option.

Aggregate statistics for each active workload partition (WPAR) in the systems

This report contains aggregate statistics for each active WPAR present in the system. This report includes following columns, one line per WPAR.

wpar: WPAR name.
nctx: Number of contexts.
pf: Number of page faults.
spf: Number of STAB faults.
pi: Number of page-ins.
err: Number of AFU errors.
exc: Number of AFU exceptions.
sig: Number of AFU signals.
stalls: Number of AFU stalls.

Flags

Flag	Description
-a	Reports system-wide aggregate statistics.
-c	Reports per context statistics.
-d	Reports aggregate AFU statistics.
-p	Reports aggregate process statistics.
-k	Reports aggregate device kernel statistics.
-t `count`	Specifies the number of times the statistics must be reported.
-i `interval`	Specifies the time interval in seconds after which the statistics must be reported.
-@ ALL	Reports aggregate statistics for each active WPAR that is present in the system.
-@ `wparname`	Reports aggregate statistics for the specified WPAR.

Examples

To report system-wide aggregate statistics, enter the following command:
```
caccelstat -a
```
To report system-wide aggregate statistics for 10 times at 1-second interval, enter the following command:
```
caccelstat -a -t 10 -i 1
```
To report all the context statistics for the CAPI-enabled processes in the system, enter the following command:
```
caccelstat -c
```
To report context statistics for the process 1234, enter the following command:
```
caccelstat -c -p 1234
```
To report context statistics for the device /dev/memcopy0, enter the following command:
```
caccelstat -c -d /dev/memcopy0
```
To report process statistics for all the CAPI-enabled processes in the system, enter the following command:
```
caccelstat -p
```
To report process statistics for the CAPI-enabled process 1234, enter the following command:
```
caccelstat -p 1234
```
To report device statistics for all the CAPI devices present in the system, enter the following command:
```
caccelstat -d
```
To report device statistics for the CAPI device /dev/memcopy0, enter the following command:
```
caccelstat -d /dev/memcopy0
```
To report device kernel statistics for all CAPI devices in the system, enter the following command:
```
caccelstat -k
```
To report device kernel statistics for CAPI device /dev/memcopy0, enter the following command:
```
caccelstat -k -d /dev/memcopy0
```
To report aggregate statistics for each active WPAR in the system, enter the following command:
```
caccelstat -@ ALL
```
To report process statistics for all the CAPI-enabled processes in the testWpar WPAR, enter the following command:
```
caccelstat -p -@ testWpar
```
To report device statistics for the CAPI device /dev/memcopy0 in the testWpar WPAR, enter the following command:
```
caccelstat -d -@ testWpar
```
To report all the context statistics for the CAPI-enabled processes in the testWpar, enter the following command:
```
caccelstat -c -@ testWpar
```