On Fri, 2017-05-05 at 18:22 +0800, 单卫 wrote: > > > 2017-05-02 23:05 GMT+08:00 Eric Dumazet <eric.duma...@gmail.com>: > > > Hi Shan > > 1) Your patch never reached netdev, because it was sent in > HTML format. > > 2) During Linus merge window, net-next is closed > > I am not really convinced that we need this with TCP > autotuning anyway. > Initial value of sk_sndbuf and sk_rcvbuf is really a hint. > > How often do you really tweak /proc/sys/net/ipv4 files in > production. > > > Again you posted your reply in HTML, so netdev did not get it.
> > Historically, most products have not adjusted any wmem/rmem > parameters. > In our experience to adjust the parameters, you can get a better visit > experience, faster. > > Now more and more products want to change this default value. > But for a product, once we adjust the parameters, it is rarely > adjusted. > > > We hope that this patch can provide a capacity: adjust the parameters > online and quickly take effect. > > Without this patch, it is time consuming for us to adjust the > parameters and need to migrate flows. > > > > > Please provide more information, like what actual values you > change back > and forth. > > > > There is no a fixed value here. big value, will take up more memory > resources. > But we are also willing to sacrifice memory for performance. > > Select the parameter value based on the characteristics of the > product. > > > For example, for 1MB files, we will set buffsize to 1611646 = (1MB / > 1460) * 2244 > Mss = 1460, SKB_TRUESIZE = 2244 > > > With TCP autotuning, reach the buffer size until the cwnd equal 359 > from 10. Looks like a problem with autotuning because you might have HZ=100 or HZ=250 maybe ? We want to make autotuning better, so that tweaks like that are no longer needed. > >
