On Fri, Jul 13, 2012 at 4:41 PM, Christoph Hellwig <hch at infradead.org> wrote: > On Wed, Jul 11, 2012 at 05:27:10PM +0800, HaiTing Yao wrote: >> I tested one 1G sized VDI I/O. >> >> Reading performance has improvment. > > Do you have profiles for both runs? I'm a bit confused about the > imrovements as all socket reads/writes should be non-blocking and not > actually use up much time in the main thread. > I tried to find profiles and got confused also. With multible threads, we can explain the improvement easy. I do not find the root reason for the improvement with only one reading request at the same time, but the improvement indeed exsited on my machines. Thanks Wujue |