On Mon, Jun 25, 2012 at 10:50:08PM +0800, levin li wrote: > Yes, it's the best way to solve this problem, I'm writing the new patch. How about sorting this out incrementall - first a series to just send all the requests to the local gateway node via TCP, then a second one to optimize this path. It would be nice if the second series could also replace the direct calls into the local I/O routines from forward_*_obj_req with that new fast path. |