a query on trade-off between tape/dasd

raghavmcs · Active User Joined: 14 Jul 2005 Posts: 105

Dear Experts,

There is one of the process in our system where 8 monthly jobs running on last business day are using the main driver file from Tape.The file size is around 2 gb.This is a flat file.We have the same file available at that time on DASD.
When I looked into the those jobs they all are just creating some files/reports which are being used by other system or for reporting purposes.
For some of the jobs I did a production test by running the process with the DASD file and I was able to to the drastic reduction in Elapsed time to CPU time ratio.
In couple of cases I was able to see the CPU time also got minimized.But some of the process came up with a minor increase in CPU time also!!

I have verified the reports/datasets from actual production run and that out of my test production run,they are same..

Need your help/advise for how I should actually link all this.Just to add some of the processes are reading around 10 VSAm files also.

I know there is lot of elapsed time which can be saved as I do have those statistics but what is happening with the CPU time I am bit confused!!
Please advise or let me know if any other information may be needed from my side,thanks

Robert Sample · Posted: Wed Jun 10, 2009 12:15 am

What is a "minor increase" in CPU time? I/O does require a small bit of CPU time, so changing from a serialized tape process to parallel disk process could raise (or lower -- depending on circumstances) the CPU time usage a bit.

raghavmcs · Active User Joined: 14 Jul 2005 Posts: 105

Robert,

One of the job I was able to see when it was reading the input file from Tape the CPU time was .14 and when I ran it with DASD file it was .15.

I would like to update that all of the 8 monthly jobs will not be using th DASD file at the same time,they all run at different-2 times.

Could you please elaborate your sentence so changing from a serialized tape process to parallel disk process could raise (or lower -- depending on circumstances) the CPU time usage a bit.

I am curious to know could be known circumstances in your mind as of now,thanks

Robert Sample · Posted: Wed Jun 10, 2009 1:18 am

First of all, a difference of .14 to .15 could be no difference -- .144999 versus .145001 for example. Unless you went to the trouble of pulling the SMF data for the jobs and computed all the digits of the CPU time usage, you can't really say those numbers are different.

Second, tape jobs run in sequence -- they are serialized because two jobs can't be reading different parts of the tape at the same time -- the operating system won't allow that.

Third, disk jobs, unless serialized on purpose via DISP=OLD, can run in parallel -- multiple jobs can be reading different parts of the file at the same time.

Factors to consider include number of buffers, channel contention, disk contention, number of jobs running simultaneously, amount of processing versus I/O being done, buffer sizes, and on and on and on. Without running tests you can't really say that given jobs running against tape will run more or less CPU time than the same jobs running against disk. I would not expect to see large differences but a percent or two may be possible. And as with all performance measurements, looking at one single job against another single job is not a very good indicator of how things will run over time.

If you're concerned about all the jobs hitting the same file at the same time, you could schedule some of them for one time, the rest for a different time.

raghavmcs · Active User Joined: 14 Jul 2005 Posts: 105

thanks Robert,all this information is really helpfull,I read the whole with more wide eyes and understand it...