Databases
cancel
Showing results for 
Search instead for 
Did you mean: 

Counting physical I/O for processes?

SOLVED
Go to solution
Mott Given
Frequent Advisor

Counting physical I/O for processes?

We use a third party tool, BMC's Patrol for UNIX Perform/Predict which has its own data collector to monitor our HP-UX systems. I am trying to figure out how much the product is guessing when it comes to allocating I/O activity on a process basis to each disk. It does not look like the basic HP-UX tools like iostat, ps, etc. give the I/O for a process to each disk so I figure the BMC product must be guessing. Is my assumption true or false?
Mott Given, mgiven@dsdc.dla.mil
4 REPLIES
John Carr_2
Honored Contributor

Re: Counting physical I/O for processes?

Hi

If BMC was simply guessing then I doubt if it would have sold such a large number of copies.
The predictive side is based on stats generated over a period of time.

John.
Steven Gillard_2
Honored Contributor
Solution

Re: Counting physical I/O for processes?

No, it wouldn't be guessing. This information usually comes from the pstat*(2) family of system calls. For details see the pstat man page, and have a look at the /usr/include/sys/pstat/*.h files for details on the stats that are available.

And of course theres the measureware agent which collects and logs these stats over time. If you're running this the extract command can be used to export the data into text format so it can be viewed.

Regards,
Steve
Mott Given
Frequent Advisor

Re: Counting physical I/O for processes?

The reason that I was thinking that BMC's Patrol for UNIX Perform/Predict might be guessing is that when you browse the text of a model file they have built:
1) for a given disk device, each workload has exactly the same # of I/O's to that device
per transaction
2) for each workload they are assuming 50% reads and 50% writes, and are assuming that all I/O's are synchronous. It seems that if they knew the
I/O's to each disk for a process that they would not have to guess about what %'s
are reads, writes, and asynch or synchronous.
3) They use their own data collector and are not getting
data from Measureware.
4) On a per process basis they seem to be measuring blocks in and out instead of I/O's.
BKUMAR
Frequent Advisor

Re: Counting physical I/O for processes?

Hi

As steve mentioned.....it can utilize the hp-ux standard performance commands.......instead of measureware ......you can check by your own samplings

of sar,vmstat,iostat,pstat,netstat -an (for with bmc stats....it shouldnt have guess....

keep monitoring with unix commands and compare with bmc....

Unix Administration Most Dedicated and Challenging Career in the World