Re: Memory fragmentation?

On Nov 21, 2007 12:27 AM, Justin Dolske <[email protected]> wrote:
> My assumption here was that the VM system can more efficiently deal with
> large contiguous regions of a process's address space than when it's
> scattered with unmapped holes. To me, "I'm using 0x10000000 to
> 0x20000000" implies simple (fast) accounting, while adding hundreds or
> thousands of gaps to that would increase overhead.

That sort of address space fragmentation (and accounting overhead on
task switch or other map-related activities) can indeed be a concern,
and was something we needed to watch out for when doing high
performance I/O work in the Linux kernel.  That was several years ago,
but I share your skepticism, and would want to see some high-fidelity
measurements and modelling before recommending that we start punching
holes in our mappings willy-nilly.

Mike