zTPF: Change system dumps to write out in at least 1MB frames

[jjohnst-us]

To improve performance of system dumps being written out on z/TPF the size
of the chunks before a write is issued was changed from 4k to 1MB.

The change dramatically reduces the number of I/Os when writing system
dumps to the file system.

[skip ci]

[babsingh]

These changes will have to be merged without any Jenkins testing since no ZTPF machines exist in the build farm. @jjohnst-us

• Can you confirm if you have tested these changes at your end?
• Can you provide an estimate for the reduction in the number of I/Os?
• Was there any impact on footprint?

[jjohnst-us]

@babsingh
Yes, the change has been in our system for a while, we are just getting around to doing the submission.
The chunk size never should have been set to 4k, given that system core files can be hundreds of MBs.
I don't recall specifically the I/O reduction. However, we can think of each 4k 'write' driving an I/O,
as opposed to batching 1MB worth of system dump data before driving an I/O. This would reduce
the number of request per 1MB of data to dump by roughly by 250. I due remember the wall clock time
to write out a large dump was reduced by 40%-60%.
I wasn't measuring footprint but we haven't had any reports of any changes in footprint.

Correct we currently don't have z/TPF machines in the build farm.