RE: [patch] speed up single bio_vec allocation

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



Jens Axboe wrote on Monday, December 04, 2006 12:07 PM
> On Mon, Dec 04 2006, Chen, Kenneth W wrote:
> > On 64-bit arch like x86_64, struct bio is 104 byte.  Since bio slab is
> > created with SLAB_HWCACHE_ALIGN flag, there are usually spare memory
> > available at the end of bio.  I think we can utilize that memory for
> > bio_vec allocation.  The purpose is not so much on saving memory consumption
> > for bio_vec, instead, I'm attempting to optimize away a call to bvec_alloc_bs.
> > 
> > So here is a patch to do just that for 1 segment bio_vec (we currently only
> > have space for 1 on 2.6.19).  And the detection whether there are spare space
> > available is dynamically calculated at compile time.  If there are no space
> > available, there will be no run time cost at all because gcc simply optimize
> > away all the code added in this patch.  If there are space available, the only
> > run time check is to see what the size of iovec is and we do appropriate
> > assignment to bio->bi_io_Vec etc.  The cost is minimal and we gain a whole
> > lot back from not calling bvec_alloc_bs() function.
> > 
> > I tried to use cache_line_size() to find out the alignment of struct bio, but
> > stumbled on that it is a runtime function for x86_64. So instead I made bio
> > to hint to the slab allocator to align on 32 byte (slab will use the larger
> > value of hw cache line and caller hints of "align").  I think it is a sane
> > number for majority of the CPUs out in the world.
> 
> Any benchmarks for this one?

About 0.2% on database transaction processing benchmark.  It was done a while
back on top of a major Linux vendor kernel. I will retest it again for 2.6.19.


> [...]
> 
> Another idea would be to kill SLAB_HWCACHE_ALIGN (it's pretty pointless,
> I bet), and always alloc sizeof(*bio) + sizeof(*bvl) in one go when a
> bio is allocated. It doesn't add a lot of overhead even for the case
> where we do > 1 page bios, and it gets rid of the dual allocation for
> the 1 page bio.

I will try that too.  I'm a bit touchy about sharing a cache line for different
bio.  But given that there are 200,000 I/O per second we are currently pushing
the kernel, the chances of two cpu working on two bio that sits in the same
cache line are pretty small.

- Ken
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to [email protected]
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

[Index of Archives]     [Kernel Newbies]     [Netfilter]     [Bugtraq]     [Photo]     [Stuff]     [Gimp]     [Yosemite News]     [MIPS Linux]     [ARM Linux]     [Linux Security]     [Linux RAID]     [Video 4 Linux]     [Linux for the blind]     [Linux Resources]
  Powered by Linux