On Tue, Nov 15, 2011 at 09:31:11PM -0500, tao.peng at emc.com wrote: > With buffer IO and delayed allocation, I agree it can see the whole > object. But DIO... do you plan to buffer the whole object in memory > and send it down in one write system call? With current 4MB objects, > it seems reasonable. But how about future large objects? The current code always reads and writes full object size chunks in store_create_and_write_obj_cow. If sheepdog switches to partial object size COWs using fallocate would probably be beneficial. |