[lwip-users] RTO calculated to 0 msec ??? (follow-up)

[Bonny Gijzen]

Hi again,

 

 

After examining the log some more I also noticed that the RTT measurement behaves incorrectly.

Mostly I get a string like this in my log:

 

tcp_receive: experienced rtt 0 ticks (0 msec).

or

tcp_receive: experienced rtt 1 ticks (500 msec).

or

tcp_receive: experienced rtt 2 ticks (1000 msec).

 

 

But after 1 day of running (or so) I see this:

 

tcp_receive: experienced rtt 65537 ticks (32768500 msec).

 

And it never gets back again to a LOW value.

It seems there is an offset of 65536 added to the measured RTT value.

I have examined the tcp_in.c code and noticed this:

m = tcp_ticks - pcb->rttest;

 

So when the segment is sent away the ->rttest value is set to tcp_ticks,

and on receiption of the ACK the difference of ->rttest and the current tcp_ticks value is calculated and stored in m. (=response time, expressed in timerticks (=500msec))

 

There are 2 things which strike me:

 

1) How is WRAP AROUND handled?

2) All 3 variables are differently defined:

int       m;

u32_t   tcp_ticks;

u16_t   rttest;

 

I think the offset of 65536 is directly related to the above formula and the definition of the variables themselves. I will add some more debug code and log again to see exactly what happens.

Can somebody have a look at this and perhaps explain a little of it?

 

Thanks

Bonny