From: Aviv Greenberg <avivgnet@xxxxxxxxx> Date: Mon, 4 Jul 2005 16:33:56 +0300 > This is weird since all the framework is there to enable zero > copy in sendmsg - by marking the user pages as COW until the > pskb is freed. > > There must be a reason why it's not there (the patches > are 4 years old). can someone shed some light on this? > Was this ever tried? What am i missing? It costs more to lock down the user mappings than the gain you get from the zero-copy transfer. Also, the user almost always modifies the buffer right after the sendmsg() call so the gain is usually nil. Benchmarks that just send off sendmsg() calls on the same buffer over and over and never modify that buffer are far from what happens in real life. In the end analysis, it's a total lose, just ask the SGI folks who did this in IRIX many moons ago. There are so many applications and benchmarks out that that have IRIX ifdefs in them to modify the buffer handling in the application in order to avoid all the COW'ing faults from the sendmsg() zerocopy stuff. - : send the line "unsubscribe linux-net" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
