On Fri, Feb 5, 2016 at 6:33 AM, Sage Weil <sweil@xxxxxxxxxx> wrote: > https://github.com/ceph/ceph/pull/5534 > > does two things: (1) combine the buffer::raw reference count and buffer > into a single allocation (ala make_shared), and (2) convert buffer::list > into an intrusive_list, while embedding N (=3) buffer::ptr's in the > buffer::raw so that most buffers require only 1 allocation (provided they > aren't on more than N lists). > > This needs a rebase, and some benchmarking. But before I pull it up, is > this the right direction? I saw little benefit from (2) after doing > (1), but I was doing a really simple microbenchmark. Do we want both, or > just 1, or is there an alternative we should consider? I think a another way is separating the owner of memory and the reference to the part of memory. Like Buffer structure in RAMCloud(https://github.com/PlatformLab/RAMCloud/blob/master/src/Buffer.h) 1. The Buffer(like bufferlist) own "vector<void*> allocations" to store all associated memory to keep them. It will be convenient for addition malloc(like getRange) 2. Chunk(like ptr) maintain *next ptr instead std::list Hmm, it doesn't have much difference. FYI > > sage > -- > To unsubscribe from this list: send the line "unsubscribe ceph-devel" 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 ceph-devel" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
