Content-Type: multipart/mixed; boundary="===============8808796094386900552==" MIME-Version: 1.0 From: Florian Westphal To: mptcp at lists.01.org Subject: [MPTCP] [PATCH mptcp-next 0/2] mptcp: add receive buffer auto-tuning Date: Thu, 28 May 2020 15:02:37 +0200 Message-ID: <20200528130239.16498-1-fw@strlen.de> X-Status: X-Keywords: X-UID: 4522 --===============8808796094386900552== Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: quoted-printable This patch adds rx buffer auto-tuning. In oder to better test this, the first patch extends the test script so it can be run multiple times with identical file size. Furthermore, the new '-t' switch allows to always to tcp fallback and even tcp <-> tcp tests. The expectation here is that there is little to no difference in transfer time between mptcp <-> mptcp and tcp-only transfers. Without the second patch, there is significant difference when delay is used due to the missing rx autotuning. The new mptcp tune function is very similar to the tcp one, except that it uses the highest rtt of any subflow. Its also needed to adjust the subflow rcvbuf: if that is not done, there are backlog drops at the subflow level (default rx buffer on my vm is only 128kb, which doesn't provide enough 'cushion' until work queue can drain the ssk queue). --===============8808796094386900552==--