Dan writes:
> Even when someone runs query and flushes
> everyone's CI out of ram and kills the machine.
...
> BTW does anyone notice store/restore takes the piss out of your machine?
No, purge does that :) (purge piss.pub.sys) (ouch)
Seriously, with program like backup (STORE), an operating system designer,
a tool designer, and the user need to ask themselves the question:
Should this program run as fast as possible, even if that interferes
with the performance of the other applications on the system?
This is an impossible question to accurately answer without knowing
everything about everything. For example, what if the answer is "yes",
and the user starts *two* such programs on the machine at the same time?
On the Classic, I answered the above question for one tool with the answer:
Sort of...I'm willing to take up to 1/2 the physical memory, and
1/2 the available DST entries, but no more.
That represented an informed attempt at balancing the desire for performance
with the desire to be a good "computer citizen". Note that my solution
meant that in the middle of the night, with no other users logged on, I'd
be underutilizing the machine ... on the other hand, if I said "wow, no
one is here...grab every available DST", then when the operator tried to
run a program 2 seconds later, he/she would be unable to!
With MPE/iX, we lack the ability to give MPE guidance about our anticipated
minimum/maximum resource usage needs. Some sophisticated (and privileged!)
programs attempt to avoid the system impact Dan describes by doing things
like:
recipe for serially reading a large file:
FLUSH prior pages (the ones we are done with)
PREFETCH next pages
handle current pages
Note that without an explicit FLUSH of some kind, MPE's only clue that
you might not be needing a page in memory again is the Memory Manager's
"how recently was this page used" logic ... which isn't as knowledgable about
the application as the application programmer should be.
--
Stan Sieler [log in to unmask]
http://www.allegro.com/sieler.html
|
