a-lunev commented on a change in pull request #5239:
URL: https://github.com/apache/incubator-nuttx/pull/5239#discussion_r785444768
##########
File path: net/tcp/tcp_sendfile.c
##########
@@ -495,6 +495,9 @@ ssize_t tcp_sendfile(FAR struct socket *psock, FAR struct
file *infile,
*/
net_lock();
+#ifdef CONFIG_NET_TCP_WRITE_BUFFERS
+ conn->sendfile = true;
Review comment:
I have the following plan concerning TCP rework:
1. Fix already found issues in tcp_sendfile. One of the fix is the current
PR #5239.
2. Also I have almost prepared the next patch that fixes another issue in
tcp_sendfile (+ code optimization), and at the same time the TCP payload
throughput of tcp_sendfile is increased from 234 Mbits/s up to about 432 Mbit/s
(sim:tcpblaster config on my Linux setup).
3. Fix the rest of already found issues in tcp_sendfile.
4. Try to modify tcp_send_unbuffered.c to achieve similar TCP payload
throughput as already achieved for tcp_sendfile.
5. Try to modify tcp_send_buffered.c to achieve similar TCP payload
throughput as already achieved for tcp_sendfile.
6. If TCP payload throughput for tcp_send_buffered.c is achieved at least as
tcp_sendfile already shows, then I could try to simulate tcp_sendfile over
tcp_send_buffered.c (CONFIG_NET_TCP_WRITE_BUFFERS=y) and check if the
throughput is not reduced because of the simulation.
Right now in case of sim:tcpblaster config the TCP throughput of
tcp_send_buffered.c is only 36.7 Kbit/s. If I try to simulate tcp_sendfile over
tcp_send_buffered.c right now, the throughput of tcp_sendfile will not be
higher 36.7 Kbit/s for sure.
Thus I'm suggesting to go step by step. I'm going to prepare the patches as
I outlined.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
