Re: Active Memory Defragmentation: Our implementation & problems

[Timothy Miller]

Richard B. Johnson wrote:

> If this is an Intel x86 machine, it is impossible for pages
> to get fragmented in the first place. The hardware allows any
> page, from anywhere in memory, to be concatenated into linear
> virtual address space. Even the kernel address space is virtual.
> The only time you need physically-adjacent pages is if you
> are doing DMA that is more than a page-length at a time. The
> kernel keeps a bunch of those pages around for just that
> purpose.
>
> So, if you are making a "memory defragmenter", it is a CPU time-sink.
> That's all.

Would memory fragmentation have any appreciable impact on L2 cache line 
collisions?

Would defragmenting it help?

In the case of the Opteron, there is a 1M cache that is (I forget) N-way 
set associative, and it's physically indexed.  If a bunch of pages were 
located such that there were a disproportionately large number of lines 
which hit the same tag, you could be thrashing the cache.

There are two ways to deal with this:  (1) intelligently locates pages 
in physical memory; (2) hope that natural entropy keeps things random 
enough that it doesn't matter.


-
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/