Re: [PATCH] KSM: Replace jhash2 with xxhash

[Andi Kleen <ak@linux.intel.com>]

On Fri, Sep 22, 2017 at 12:37:25AM +0300, Timofey Titovets wrote:
> With defaults:
> jhash2: ~4.7%
> xxhash64: ~3.3%
> 
> 3.3/4.7 ~= 0.7 -> Profit: ~30%
> 11/18   ~= 0.6 -> Profit: ~40%
> (if i calculate correctly of course)

Sounds good.

Please add all performance information to the changelog.

> 
> >> >> @@ -51,6 +52,12 @@
> >> >>  #define DO_NUMA(x)   do { } while (0)
> >> >>  #endif
> >> >>
> >> >> +#if BITS_PER_LONG == 64
> >> >> +typedef      u64     xxhash;
> >> >> +#else
> >> >> +typedef      u32     xxhash;
> >> >> +#endif
> >> >
> >> > This should be in xxhash.h ?
> >>
> >> This is a "hack", for compile time chose appropriate hash function.
> >> xxhash ported from upstream code,
> >> upstream version don't do that (IMHO), as this useless in most cases.
> >> That only can be useful for memory only hashes.
> >> Because for persistent data it's obvious to always use one hash type 32/64.
> >
> > I don't think it's a hack. It makes sense. Just should be done centrally
> > in Linux, not in a specific user.
> 
> So, i must add separate patch for xxhash.h?

Yes.

> If yes, may be you can suggest which list must be in copy?
> (i can't find any info about maintainers of ./lib/ in MAINTAINERS)

Just copy linux-kernel. It would be all merged together.

> If we decide to patch xxhash.h,
> may be that will be better to wrap above if-else by something like:
> /*
>  * Only for in memory use
>  */
> xxhash_t xxhash(const void *input, size_t length, uint64_t seed);

Yes that's fine.

-Andi

--
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>