On Sun, 23 Jun 2002 12:56:39 -0700 (PDT)
[EMAIL PROTECTED] (dfox) wrote:

> 
> > fragmentation in memory pages.. the same little sketches i made 
> > would apply, but change the "file 1" and "file 2" with "process 1"
> > and process 2"  )
> 
> I doubt that's a serious issue. It's kind of silly, really, since any
> portion of RAM is just as quickly issued as any other. But this is
> virtual memory (modern OSes) we're discussing. And virtual memory
> is viewed simplistically as having some memory on some other media
> (i.e., a swap partition). That is only one aspect of it. In reality,
> virtual memory means that a process residing at some address N need
> not really reside at address N on the RAM chips. Realisically, N is
> only an offset which is maintained by the processor, with the OSes
> help. One might be able to find out where a process resides physically
> in the RAM if it starts at address 10000 and continues to address
> 20000, but it would be kind of pointless. And, if the process is 
> larger than a page of memory (4K) then one page might be located 
> way away from the others. In fact, in a continually running system,
> it's practically guaranteed that this is the case.
>

actually, external fragmentation of the memory CAN be a problem.
yes, running a fragmented process takes the same time, but it 
bring other problems as well, to the memory management system.

it's been a long time since i've studied the issue, and i can't
remember exactly what was the overhead involved, but it had 
something to do with the operating system's process tables'
management.

in fact, now that you metion swap, fragmenting the memory space will
probably mean fragmentation of the swap space in very short time.  both
external and internal fragmentation in the swap space can degrade
performance seriously, by forcing the HD to change the head's position
more often, it can result in slow swapping in and out of memory pages.

there are many aspects involved in swap performance, one of them
being that swap is "an extension of the phisical memory" but only as a 
"storage" space. no process code can be run from the swap. only
"blocked" or "waiting" process pages will be swapped out, and when they
need to change state to "ready, running", they must be loaded into the
RAM. and this is when fragmentation begins to be a problem...



Damian




Want to buy your Pack or Services from MandrakeSoft? 
Go to http://www.mandrakestore.com

Reply via email to