On Thu, Dec 07 2006, Nate Diller wrote: > On 12/7/06, Chen, Kenneth W <[EMAIL PROTECTED]> wrote: > >Nate Diller wrote on Thursday, December 07, 2006 1:46 PM > >> the current code is straightforward and obviously correct. you want > >> to make the alloc/dealloc paths more complex, by special-casing for an > >> arbitrary limit of "small" I/O, AFAICT. of *course* you can expect > >> less overhead when you're doing one large I/O vs. two small ones, > >> that's the whole reason we have all this code to try to coalesce > >> contiguous I/O, do readahead, swap page clustering, etc. we *want* > >> more complexity if it will get us bigger I/Os. I don't see why we > >> want more complexity to reduce the *inherent* penalty of doing smaller > >> ones. > > > >You should check out the latest proposal from Jens Axboe which treats > >all biovec size the same and stuff it along with struct bio. I think > >it is a better approach than my first cut of special casing 1 segment > >biovec. His patch will speed up all sized I/O. > > i rather agree with his reservations on that, since we'd be making the > allocator's job harder by requesting order 1 pages for all allocations > on x86_64 large I/O patterns. but it reduces complexity instead of > increasing it ... can you produce some benchmarks not just for your > workload but for one that triggers the order 1 case? biovec-(256) > transfers are more common than you seem to think, and if the allocator > can't do it, that forces the bio code to fall back to 2 x biovec-128, > which, as you indicated above, would show a real penalty.
The question is if the slab allocator is only doing 2^0 order allocations for the 256-page bio_vec currently - it's at 4096 bytes, so potentially (I suspect) the worst size it could be. On the 1 vs many page bio_vec patterns, I agree with Nate. I do see lots of larger bio_vecs here. > 1 page bio_vec usage is also becoming more prevalent, not less. So optimizing for a benchmark case that predominately uses 1 page bio's is indeed a silly thing. -- Jens Axboe - 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/
