Re: [lwip-users] speed and delay question with uc/os-ii

[jarkko . loima]

Thanks for quick answer.

I have to send either (4260) Bytes once in 100mS or 73 Bytes in every 2mSec...

Buffering & sending data with TCP is more reliable way but it takes about 20mSec
to send that packet. Reason for problems is that i'm using interrupt driven
system and sending data through TCP overrides all lower level interrupts. If i
chabge prioritys of interrupts incoming data is ok but TCP-start's sending out
of order & retransmissions...

So what would you suggest? Do you think that i could send UDP-packet once in
every 2mSec?

Other end is w2k Visual C++, program with casynckSocket...

-JLoima

Quoting "K.J. Mansley" <address@hidden>:

> On Wed, 2004-08-25 at 09:13, address@hidden wrote:
> > Hi!
> >
> > I'm using lwip with uc/os-ii. I'd like to know which is the wastest &
> delayless
> > way to send data (size of data to send is about 4260 Bytes).
> >
> > I'm using TCP with netconn_write() at the moment but have a little problems
> with
> > delay (20 ms) & retransmissions...
> >
> > So, is UPD faster way or should i use tcp_write() or something like that???
> I
> > think that with UDP i could get rid of these retransmissions.
>
> The raw API (e.g. tcp_write() etc) will be faster than using netconn_*
>
> UDP may make a small difference, as it's a simpler protocol, but if
> you're losing data you'll need to do retransmissions whatever transport
> protocol you choose.  I would try to work out why data need to be
> retransmitted, and address that.  If you are just doing a single short
> transfer over each connection you may find UDP helps by not having the
> "slow start" phase of TCP.
>
> Hope that helps,
>
> Kieran
>
>
>
> _______________________________________________
> lwip-users mailing list
> address@hidden
> http://lists.nongnu.org/mailman/listinfo/lwip-users
>