Thanks to the following:
Jane.Kramer_at_oberlin.edu
A.Mahendra.Rajah_at_Meena.CC.URegina.CA
tpb_at_doctor.zk3.dec.com
alphaosf_at_ini.dec.com
alan_at_nabeth.cxo.dec.com
george_at_vache.ucdavis.edu
You guys are great! You're explainations were MUCH clearer than our Gold
Support Team! Here's just one of the greatly detailed messages I received
explaining the ins and outs of "eager" and "lazy" swap modes:
On Fri, 4 Dec 1998 alan_at_nabeth.cxo.dec.com wrote:
>
> In the default mode of page/swap space reservation, each
> page of virtual memory that is created and *may* need
> page/swap space *allocated* for it, a page is reserved.
> For example an program doing malloc(3) will need pages
> of memory that have to be swapped or paged out eventually;
> this needs pages reserved. A text page that is read in
> from an a.out may not need space reserved because it can
> be read from the file system. A page that is created using
> mmap and backed by a file won't need space reserved, since
> the file acts as the page/swap space. I'm not sure how
> text is actually handled, so mmap and a file is better
> example of virtual memory created and not needing system
> page/swap space.
>
> There are two ways to reduce the percentage of reserved
> space:
>
> 1. Add more space (thus raising the amount available and
> reducing the percentage reserved).
>
> 2. Use less memory.
>
> An alternative to reserving virtual memory up front is
> not to (gotta love binary choices...). To switch modes
> remove the symbolic link /sbin/swapdefault and reboot.
>
> The disadvantage of this mode (AND THIS IS IMPORTANT) is
> that allows a dead lock situation where a process needs
> page/swap space and there isn't any. The system will
> prevent the dead lock from occuring by killing processes
> until there is sufficient page/swap space. It seems to
> prefer idle processes, but in extreme cases will take
> whatever it can find. It ignores little.
Cory
Received on Fri Dec 04 1998 - 23:31:24 NZDT