Selon Adi Oanca <adioanca@xxxxxxxxxxxxxx>: > Axel Dörfler wrote: > > > > > > As I said, the buffer size has *nothing* to do with this anymore. > > Have a look at LinkMsgSender::StartMessage() maybe you can believe me > > then :-) > > I had a look yesterday, that's why I did that test in the first > place. The buffer always grows at a max of 64K, it never shrinks back to > 2K. > Have another look, and this time maybe you should put you glasses. :-P > > I cannot believe you as I have my tests a a proof. ;-) > Looking at the code would theorically make me think Axel is right : the buffer size is always increased, not decreased. But as flushs are done on a hard watermark, your tests shouldn't show increased latency because of a high buffer size (unless a bug :) ). Jérôme