On Fri, Jan 13, 2017 at 09:13:42AM -0800, Christoph Hellwig wrote: > On Wed, Jan 11, 2017 at 12:54:04PM -0500, Brian Foster wrote: > > Hi all, > > > > Here's v2 of the cow fork speculative prealloc patches. No major changes > > here, as outlined below. Thoughts on any of this? > > Any results? Better performance, less extents for certain workloads? Sorry, I neglected to pull that forward from the rfc. That discussion is here: http://www.spinics.net/lists/linux-xfs/msg02152.html I should probably repeat some of this testing at this point, but the general result is a reduction in extent count for parallel cow fork allocator workloads. E.g., attempting to address the same problem of parallel data fork allocators competing for contiguous space, resolved by traditional speculative preallocation. I was also kind of thinking about whether this could work around the need to have a default cowextsz hint at all, allowing that feature to be used more like the traditional extent size hint as well. This is distinctly more aggressive for use cases like large vm image files, however. At the same time, the idea is to make it transparent enough that it shouldn't really affect the user (i.e., via throttling, background and on demand reclaim, also just like normal preallocation). Brian > -- > To unsubscribe from this list: send the line "unsubscribe linux-xfs" in > the body of a message to majordomo@xxxxxxxxxxxxxxx > More majordomo info at http://vger.kernel.org/majordomo-info.html -- To unsubscribe from this list: send the line "unsubscribe linux-xfs" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
