Re: [PATCH] 2.2.17pre7 VM enhancement Re: I/O performance on

[Marcelo Tosatti]

On Wed, 28 Jun 2000, Andrea Arcangeli wrote:

> On Wed, 28 Jun 2000, Marcelo Tosatti wrote:
> 
> >Innocent processes (e.g. syslogd, klogd, bash) are getting killed
> >with this patches. 
> 
> While you are oom, or while you're running mmap002, right? 
Yep. 

> If so that's
> intentional and I'm very aware of the MAP_SHARED problem but what we did
> so far to fix that MAP_SHARED issue is band-aid (that could hurt the other
> cases that works right). 

With "right cases" here you mean waiting for IO completion while there is
freeable pages around?
I suggested a fix for that in the past email.

Do you see any problem with this fix? If not, i'm going to send a patch
soon. 

> I believe that the real problem of MAP_SHARED is
> that we don't know anything about the state of the pages in MAP_SHARED
> segment. We consider all them as clean without doing any write throttling
> and we allow the VM to become fully dirty; while for good reasons in the
> rest of the kernel we instead assume that there is a max percentage of the
> RAM that could become dirty at the same time.

Ok. Are you going to fix that? 

> Then there is the fact that the VM isn't handling gracefully the case
> where all the memory is dirty, that's another issue and it's secondary
> problem IMHO (all the VM shouldn't become dirty so easily as now in first
> place).