measureware extract utility v/s sar,vmstat,iostat

[Deepak Seth_1]

Iam leaning the use of measureware extract command . It seems much simplier and faster as compare to using perf view. I am looking for develping some scripts which can give me summaries of disk,cpu,memory and network usuage for a given period of time . Can anybody sugest me critical metrics (like GBL_CPU_TOTAL_UTIL etc ..) to look for on a regualr basis for the above category(disk,cpu,memory,network). My intial experiment to find the buffer cache doesn't yielded good results. I ran sar -b for 4 hrs and collected the data . Then i try using the measureware extract to analysise the data for the same 4 hrs. for the GBL_DISK_VM_READ_RATE , GBL_DISK_VM_READ metrics. The measureware extract brings a different output? What should i use in long run for my monitoring scripts - Measureware or Sar,Vmstat etc


Any direction or suggestions ?

[G. Vrijhoeven]

Hi,

It all depents on haw mutch $ you want to, or can spent and who wants the information.. explane sar output to a manager or draw a graph with maesureware.

Measureware v/s Sar, it all depents on the number off times samples are collected and the interval. I was told that measureware is more trust able but they both do snap shots from the kernel so i am not sure.

hope this will help a bit.

[Craig Rants]

measureware will give you more accurate information compared with sar, vmstat, etc...

measureware gathers its information through daemons, whereas sar and the like take samples.

Depends on how detailed you want to get.

You will probably be able to write scripts with sar and vmstat more easily, but some experience with measureware and you can do much more...

Enjoy your scripting.
C

[Roger Baptiste]

<>

Each has its place. Measureware is generally used
for collecting performance data and looking at it for
an analysis point of view.

Sar, vmstat -> we use for
a quick and immediate use.
To see the cpu load distribution etc.
I need to add that, i run
scripts on certain options
of sar (sar -d for instance
to gather data on disk i/o.).

I would say, each tool complements one another.
measureware is the reliable
of the lot, since iirr, it
reads from the kernel registers. ( i don''t rememember the details).

What params to monitor using
extract? - make copies
of /var/opt/perf/reptall file
and comment out different key values as per your needs.
(GBL_CPU_TOTAL_UTIL etc.. they
are self-explanatory).


HTH
raj

[Deepak Seth_1]

Coming back to the key question. What will be those really good metrics to include in your monitoring scripts ? I just want names ?

Here is what i have for Global

............... Global CPU Metrics

* DATE
GBL_CPU_SYS_MODE_UTIL
GBL_CPU_USER_MODE_UTIL
GBL_PRI_QUEUE
BLANK
*
................ GlobalDisk Metrics
*
DATE
GBL_DISK_UTIL_PEAK
GBL_DISK_VM_READ_RATE
GBL_DISK_VM_READ

............. Global Memory Metrics
*
* DATE
GBL_MEM_UTIL
GBL_MEM_USER_UTIL
GBL_MEM_PAGEOUT_RATE
.......... Global Network Metrics
*
GBL_NET_IN_PACKET_RATE
GBL_NET_OUT_PACKET_RATE
GBL_NET_COLLISION_PCT

Any inputs with your experience of using measureware extract command.

[Uday_S_Ankolekar]

Hi,

You can check this directory for a better understandings
/opt/perf/examples/adviser
Goodluck...
-USA...

[Deepak Seth_1]

I think i present my question wrongly. I very well understand the use of measureware v/s sar,vmstat etc. Iam planning to build some script for my help desk which they can use to find the bottleneck. In order to show the bottleneck , i need to collect data. The option is between using the existing measureware data or running vmstat,sar to capture data . Then builing some script to show the help desk on the last 1 hrs of performance etc. Something which can be useful to take some decision.

Did any of you have similar scripts to share or any guidlines on my project.

[Roger Baptiste]

hi,

for a quick snapshot i use
these parameters:
***
GBL_CPU_TOTAL_UTIL
GBL_MEM_UTIL
GBL_SWAP_SPACE_UTIL
GBL_MEM_PAGEOUT_RATE
GBL_NUM_USER
GBL_PRI_QUEUE
GBL_RUN_QUEUE
GBL_ACTIVE_PROC
****

and run the command
extract -xp -v -gp -r /tmp/reptall -b 01/xx/yy
This will create two
files in the current directory
for process(p) and global(g)
in a table format.

Based on the values, if there
is a problem, based on whether
it is disk,cpu,or i/o , i
enable other parameters to
get more detail.

-raj