On Wed, 14 Jun 2000, Andrea Arcangeli wrote:
> On Wed, 14 Jun 2000, Rik van Riel wrote:
>
> >but since dozens of people complained about 2.2.15 VM performance
> >I guess things didn't exactly work out as expected...
>
> If an innocent task was getting the trashing mem bit set it could decrease
> interactive performance, I know. But how can it hurt stability?
Good to see that we can agree on that point. ;)
> >Indeed. It _did_ get killed, however, because try_to_free_pages()
> >in 2.2 can sometimes fail. By starting to call try_to_free_pages()
>
> It was the flushcount in vmscan.c or the free_before_allocate
> that can have made differences. I think the vmscan.c changes are
> very right. The free_before_allocate just avoids the machine to
> runs too near to the min limit but it shouldn't be necessary and
It _is_ necessary, but not for the reason you think.
If one application is sleeping in try_to_free_pages (doing
IO or somesuch), we must make sure that other programs in
the system do not just allocate all the memory that the
process in try_to_free_pages has just freed.
Ingo Molnar has done some traces to figure out where Linux
VM went wrong and observed this situation. I'm very happy
about that because this "race" is weird enough that nobody
had come up with it by just sitting and thinking about the
code ;)
regards,
Rik
-- The Internet is not a network of computers. It is a network of people. That is its real strength.Wanna talk about the kernel? irc.openprojects.net / #kernelnewbies http://www.conectiva.com/ http://www.surriel.com/
- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.rutgers.edu Please read the FAQ at http://www.tux.org/lkml/
This archive was generated by hypermail 2b29 : Thu Jun 15 2000 - 21:00:32 EST