I think the TSM example is what is showing the strength of the architecture.

Every computer load (mix of programs used for your application) have a
different need for memory.

My first home UNIX system was a non-virtual memory system, so
EVERYTHING stayed in memory, and on a 386 with under 1G ram was
'interesting'.

Once virtual memory was being used we 'paged'.  Read/wrote one page at
a time.  Shortly after we started using swapping to take a 'set of
pages', normally defined by a 'reasonable buffer' for the underlying
device.  This was especially true a devices were addressed by
cylinder/track/sector addressing.  When the linear addressing of
devices (a good thing) obfuscated the hardware using onboard micro
controllers, and the underlying architecture of the drives were
obfuscated further, add in larger on-device buffers that can help
buffer reads and sometimes writes, the optimization for device became
harder and less needed than before.

Now we basically do no paging, and only swap.  Memory sizes have
exploded (Thank you Dr. Moore!), we don't concern ourselves with
memory conservation, writing programs that keep loops into a few
pages, rather than touching lots of pages of memory, designing data
structures for high information density encoding, etc.

Oh well.  In summary, given the current state of machines:

1. what we have works pretty well overall, without much tweaking.
2. reducing swap rate is a good thing
3. normally some swap use means you are getting use of all your memory
(as cache, buffers, or program memory, still being used)
4. one of the easiest and cheapest upgrades to most systems (unless
you hit architectural limits) is to add more memory.
5. active use of swap on a continuing basis, should be reduced for
long term system health
6. there are some fairly easy kernel tuning procedures to determine
the handling of swap ( http://en.wikipedia.org/wiki/Swappiness and
other searches for swappiness and related items ), and various issues.
 It just takes research.  No one knob tunes everything.

This is NOT an exact science.  There are more variables involved than
most of us can handle in our heads.  The idea is to get your system to
do what you want with minimal overhead (and in my case, minimal extra
effort).  I have my swappiness set to 45 on my laptop, but I didn't
see a HUGE change, but it is supposed to be a little better for
interactive use.  Higher numbers are better for typical file server
use, where more buffer space normally helps.

Enjoy...
_______________________________________________
Tech mailing list
Tech@lists.lopsa.org
https://lists.lopsa.org/cgi-bin/mailman/listinfo/tech
This list provided by the League of Professional System Administrators
 http://lopsa.org/

Reply via email to