Filter
Top Products
Chapter 5. Additional Topics 41
– bo: blocks send to a block device (blocks/s)
System
– in: the number of interrupts per second (including clock interrupts)
– cs: number of context switches per second
CPU
– us: user time (percent of CPU time)
– sy: system time (percent of CPU time)
– id: idle time (percent of CPU time)
We are particularly interested in swap (paging) rates. Notice that the unit of measurement is
kilobytes per second.
The Linux swapping rate is meaningful (for our discussion) only in a steady-state condition
with a typical S/390 workload. Linux swapping while booting, or while starting FLEX-ES, is
not relevant. Unusual S/390 work, such as CLPA processing or very unusual disk access
patterns (affecting disk caches) might temporarily drive Linux into swapping. This is not good,
but can probably be tolerated for short periods.
5.4.2 Importance of Linux swapping
Why is the Linux swapping rate so important? A reasonable analogy is CICS paging in an
OS/390 system. A system with many TSO users might have sustained paging rates of
hundreds of pages per second (on a larger S/390) with no ill effects, but CICS on the same
system would require a paging rate close to zero. The problem is that the whole address
space (CICS, for example) is placed in wait when a page fault occurs. Placing CICS
4
in wait
causes all the CICS users serviced by that address space to wait while the page fault is
resolved. A page fault in a TSO user address space causes only that one user to wait.
FLEX-ES operation is close to the CICS analogy. A Linux page fault in a key FLEX-ES
process may cause the whole emulated S/390 instance to wait until the Linux page fault is
resolved.
Remember that a page fault in a FLEX-ES S/390 instance has a very different effect than a
page fault in Linux. For one thing, the S/390 page fault is seen only by the FLEX-ES
emulation program--it is an emulated page fault. It is handled, by OS/390, as a S/390 page
fault. If it occurs in a TSO user address space or a batch address space, it affects only that
address space. If it occurs in a CICS address space, it affects all the users of that CICS. This
is business as usual for OS/390.
The key message is that you should adjust your FLEX-ES system parameters (emulated
S/390 memory, disk caches, instruction cache) to avoid Linux swapping. Defining a smaller
emulated S/390 memory size may increase OS/390 paging. Of course, it would be nice to
avoid any paging, but OS/390 paging is much less damaging than Linux swapping and your
tradeoffs should always be in this direction. You can juggle disk cache versus instruction
cache versus S/390 memory allocations for your best performance. Simply be careful not to
push Linux into swapping.
5
4
Modern CICS systems ameliorate this situation in various ways; the description here should be regarded as
conceptual.
5
Again, we stress that “Linux paging” refers to steady-state operation after S/390 emulation is started. Linux booting
or FLEX-ES startup may cause Linux paging and we are not concerned with this temporary effect.