opensand team mailing list archive

[Julien BERNARD <jbernard@xxxxxxxxxxxxxxxxxxx>]

Hi Andrei,

Your vision of how the QoS works on OpenSAND is good but only for return 
link and I think this is a part of your problems.
Actually, we do not manage QoS on forward link but the configuration is 
really misleading. We are currently working on it.

> 1) Why do I observe different TCP behavior in to different cases:
>     -when  I  modify  only  "BE" buffer size and leave the rest buffers
>     with default values
>     -  and when I modify all buffers for all service classes by setting
>     them up to the same value.
>     Are  any  other  classes involved  while TCP flows pass
>     through  OpenSand (for instance, permanent UDP signalling traffic between ST, SAT
>     and GW is automatically marked and put into "SIG" queue?)?

This behavior seems strange and we need to try to reproduce it 
ourselves. There is no other class involved in other flows, the OpenSAND 
signaling is handled on different UDP channels and are completely aside 
user traffic.
Could you please send us your configuration files and describe your tcp 
flow (or the command line for traffic generating if you use iperf or 
equivalent) ?


> 2)  When  I send traffic in the direction from GW to ST and set up the
> buffer   size   (size_max)  of  the  "BE"  queue  on  GW  to 0, how is
> it possible that the traffic still able to pass through OpenSand with this zero buffer?
As I said above, we do not manage QoS on forward link, the service class 
and traffic category configuration part on GW  is only used for 
compatibility purpose at IP layer.
The fifo size used for GW is defined in "dvb_rcs_ncc > max_fifo" parameter.

> 3)   How   the  number  of  PVC  channels  (in  the  same  dvb_rcs_tal
> configuration  menu)  may  potentially  affect the behavior of passing
> flows?
I assume here that we are talking about return link, for forward this is 
not used.
The number of PVC channels should not affect consequently the behavior 
of flows as it will only increase the loop number for packet scheduling 
at layer 2 on terminals.

Best regards,

--
Julien BERNARD