git: kernel - vm - Change inactive target to 1/4 memory instead of 1/2 memory.

Johannes Hofmann johannes.hofmann at gmx.de
Wed Mar 24 09:34:02 PDT 2010


Matthew Dillon <dillon at apollo.backplane.com> wrote:
> :On my troublesome 2GB machine, the inactive memory is now ~= 750M, which
> :is about 1/3 of RAM, quite far from the 1/4 target.
> :
> :-- 
> :Francois Tigeot
> 
>    It's just a target for the pageout daemon, it isn't something the
>    machine tries to hold itself to.
> 
>    Guys... everyone.  Please don't try to interpret the values for
>    the active & inactive queues.  Even a good chunk of the analysis of
>    the swap space use is wrong.  There's absolutely nothing wrong with
>    the machine paging stuff out to swap.  The issue is only whether it
>    does so to the degree that later pageins effect performance.
> 
>    Avalon, for example, usually has 200-400MB of idle data swapped out
>    which it basically never touches.  It doesn't effect avalon's
>    performance.

One point is probabely whether swap is on a disk with otherwise high
IO rate. On my single disk system it is clearly a performance problem 
if program data is paged out due to filesystem activity because then
the paging in is very slow.

I can trigger this situation with
	hammer cleanup
but also with
	find / -type f | xargs md5

Maybe on avalon the swap space is on a separate disk in which
case paging back in is faster?

Cheers,
Johannes





More information about the Commits mailing list