Re: [ECOS] Facing problem in ecos network stack

[Nick Garnett <nickg at ecoscentric dot com>]

Arun <aruns_r@yahoo.com> writes:

> Hi all,
> 
> We are using eCOS on a custom board designed
> internally. The board has a ARM7 processor running at
> 66 Mhz.
> 
> Presently we are testing the IP stack with our
> driver.For one our tests we stream around 150K of UDP
> packets( each UDP packet is sent as one 20K chunk ).
> We hv observed that the IP stack drops quite a few
> packets. The number of packets dropped increase if we
> increase the UDP packet size . We notice that the
> packets seem to be correct as indicated by the
> 'ipstat'structure. We hv also noticed that for UDP
> packet sizes greater than 20K , the IP stack forwards
> either 1 UDP packet worth of data or a max of 40K
> 
> 
> We hv ensured and noticed that the ethernet driver is 
> receiving all the packets correctly and is passing
> them onto to the ip layer. 
> 

If the driver is receiving packets but they are not making it through
the stack then the most likely cause is that it is running out of
mbufs or clusters. You could try increasing the amount of memory
allocated for these and see if that makes a difference.

Also take a look at the thread that is reading the packets. If its
priority is too low then maybe it does not run often enough to pull
the packets out of the stack. Alternatively, if its priority is too
high, maybe it is not giving the stack's internal threads enough time
to process all the incoming packets. It can be a bit of a juggling act
to get the net stack working optimally.



-- 
Nick Garnett                    eCos Kernel Architect
http://www.ecoscentric.com/     The eCos and RedBoot experts


-- 
Before posting, please read the FAQ: http://sources.redhat.com/fom/ecos
and search the list archive: http://sources.redhat.com/ml/ecos-discuss