I have been comparing QNX4 Fleet performance versus TCP/IP 5.0.
When use on a local machine Fleet seems to be around 10 times faster then
TCP/IP. I was expecting a difference but not that big (i’m transfering 32k
I tried to understand the reason why/ While using monitor/msgprint to
figure out where is going all the CPU I discover that TCP/IP is using
writemsg a lot as oppose to replymx. That seems to add a lot of overhead
when dealing with large block largeur then 2k (that seems to be the size of
each writemsg). With a 64k recv() operation, it translates to around 30
writemsg, 29 context switches too many if you ask me.
I look at the doc but couldn’t find a way to change the size of the 2K block
to something bigger.