Re: [PATCH -mm] vmscan: make mapped executable pages the firstclass citizen

[Arjan van de Ven]

On Sun, 10 May 2009 16:37:33 -0400
Rik van Riel <riel@xxxxxxxxxx> wrote:

> Alan Cox wrote:
> 
> > Make your swap decisions depend upon I/O load on storage devices.
> > Make your paging decisions based upon writing and reading large
> > contiguous chunks (512K costs the same as 8K pretty much) - but you
> > already know that .
> 
> Even a 2MB chunk only takes 3x as much time to write to
> or read from disk as a 4kB page.

... if your disk rotates.
If instead it's a voltage level in a transistor... the opposite is
true... it starts to approach linear-with-size then ;-)
 
At least we know for the block device which of the two types it is
inside the kernel (ok, there's a few false positives towards rotating,
but those we could/should quirk away)

> 
> > Historically BSD tackled some of this by actually swapping
> > processes out once pressure got very high 
> 
> Our big problem today usually isn't throughput though,
> but latency - the time it takes to bring a previously
> inactive application back to life.

Could we do a chain? E.g. store which page we paged out next (for the
vma) as part of the first pageout, and then page them just right back
in? Or even have a (bitmap?) of pages that have been in memory for the
vma, and on a re-fault, look for other pages "nearby" that used to be
in but are now out ?

> 
> If we have any throughput related memory problems,
> they often seem to be due to TLB miss penalties.

TLB miss is cheap on x86. For most non-HPC workloads they
tend to be hidden by the out of order execution...

-- 
Arjan van de Ven 	Intel Open Source Technology Centre
For development, discussion and tips for power savings, 
visit http://www.lesswatts.org
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/