Re: [lwip-users] large latency on last TCP segment

[cat]

Hello

as probably noticed I have some LwIP struggles myself, but perhaps my idea
sparks another one  

My theory would be that writing 
tcp_write(40)
makes lwip use a single memory-managing structure internally. Which would
imply that it 'frees' the buffer once the entire structure is send - waits
for the last byte to be transmitted before freeing the space. If this is the
case than writing 
tcp_write(10)
tcp_write(10)
tcp_write(10)
tcp_write(10)
would do the trick, since you would be left with only last un-freed 10
during that long time and you could write in your next segment (again in
four chunks).

I think the long time could be a separate issue: lwip waits for more data to
fill up the last package (to reduce the number of packages) and gives up and
finally sends it after that 600ms. Did you consider turning off the naggle
algo?

Perhaps a solution  with the split tcp_write could help. Lwip is left with
last 10 units of data and is about to wait to fill up the package, and since
you could do the next tcp_write you would supply it, thus it would send it
sooner than waiting for the timeout...

Hope it helps in any way :)



--
View this message in context: 
http://lwip.100.n7.nabble.com/large-latency-on-last-TCP-segment-tp21227p21272.html
Sent from the lwip-users mailing list archive at Nabble.com.