[inforoots] Yachts and dynamic partioning

Mike Albaugh albaugh at perilin.com
Tue Nov 15 16:07:33 PST 2005


 Barrie Robinson wrote:

[ a bunch of stuff I concur with ] and then:

  - they probably worshipped IBM as saviours.  In the 70's I was shown
 the Data Centre of the University of Windsor by a recent computer
 science graduate.  When I asked why the huge IBM machine had to be
 'partitioned' and not run with memory dynamic allocation he responded
 that such an operation "was not technically possible".

And in a sense he was correct, if probably only by accident.
I started out in MIS and moved into embedded systems. What
they had in common was that the customer expected things
to work. This might be considered a bit "quaint" these days,
when people think of desktop systems if they think of computers
at all. Desktops share with their ancestors "workstation", "mini",
and "timesharing", the fact that users expect some number of random
mysterious failures. Thus, OSes like Unix System III can kill
a random "innocent" process when swap-space gets tight,
(some versions of?)  Linux (and others) can allocate shared
libraries in shared virtual address-space and "not worry" about
fragmentation (until the system wedges), Windows can use
a flawed timer algorithm that thrashes (and sometime crashes)
if the system is up "too long". The sort of stuff old-timers
such as myself learned about deadlock, order of reservation,
etc. is deemed "no longer relevant". And it is, on the desktop,
because statistically, Outlook is going to obey some email
and wipe you out _long_ before the various resource bugs
can do it.

Note that there is still (BIG) business in resource allocation
and management software, to shuffle resources within
"server clusters". It's just that these things only matter
to  "big" machines. In the 70's that was a mainframe.
Today it's a (minimum)  64-cpu cluster. It also can matter
in "little" machines run close to the limts of their performance.
I doubt you'll find your iPod swapping, or at least not running
well when it swaps. Even a modern PC falls off a cliff if it ever
swaps, which is why we provision them with ludicrous amounts
of RAM. Code that is never executed can't fail. :-)

  Mike 




More information about the inforoots mailing list