Re: [PATCH 4/4] compcache: documentation

[Nitin Gupta <ngupta@vflare.org>]

On 08/25/2009 10:33 PM, Balbir Singh wrote:


>> +It consists of three modules:
>> + - xvmalloc.ko: memory allocator
>
> I've seen your case for a custom allocator, but why can't we
>
> 1) Refactor slob and use it

SLOB is fundamentally a different allocator. It looked at it in detail
but could not image how can I make it suitable for the project. SLOB
really does not fit it.

> 2) Do we care about the optimizations in SLUB w.r.t. scalability in
> your module? If so.. will xvmalloc meet those requirements?
>

Scalability is desired which xvmalloc lacks in its current state. My
plan is to have a wrapper around xvmalloc that creates per-cpu pools
and leave xvmalloc core simple. Along with this, detailed profiling
needs to be done to see where the bottlenecks are in the core itself.


>
> What level of compression have you observed? Any speed trade-offs?
>

All the performance numbers can be found at:
http://code.google.com/p/compcache/wiki/Performance

I also summarized these in patch [0/4]:
http://lkml.org/lkml/2009/8/24/8

The compression ratio is highly workload dependent. On "generic" desktop
workload, stats show:
  - ~80% of pages compressing to PAGE_SIZE/2 or less.
  - ~1% incompressible pages.


For the speed part, please refer to performance numbers at link above.
It show cases where it help or hurts the performance.

Thanks,
Nitin


--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>