David Brownlee wrote:

> On Wed, 3 May 2000, Simon Burge wrote:
> 
> > This is in relation the ps argv stuff I've been looking at.  The current
> > libkvm strategy is to walk the page tables for the process, work out
> > if it's in kmem or on swap, and grab the data.  From a "process is
> > completed swapped out" POV, would it be better to keep with a similar
> > strategy to this (perhaps using something like uvm_swap_get() to fetch
> > data from swap) or just page in the pages that are needed?
> 
> 	Do ps and friends currently cause pages to be pulled in for other
> 	processes? This would seem to be suboptimal for a machine under
> 	heavy paging load...

If a process is swapped out, the argv info is read directly from the
swap device.  So in that case you could argue that there's unnecessary
demand on the buffer cache!  My question was more or less (without
saying) is it better to hit the buffer cache or normal user memory.
Roll on UBC...

Simon.