[tipc-discussion] TIPC SOCK_RDM lost messages

[Quentin A. <Quentin.Aebischer@USherbrooke.ca>]

Hello everyone,

I'm a research master degree student at the University of Sherbrooke,  
Canada and I'm currently trying to implement a TIPC module for the  
Conntrack-tools (http://conntrack-tools.netfilter.org/), which allows  
for users to implement high-available, cluster-based stateful firewalls.
My goal would be to use TIPC as protocol for the state-replication part.

So far, I've successfully implemented and tested unicast TIPC sockets  
that communicates in a connectionless yet reliable manner (using  
SOCK_RDM), using functionnal addressing (names that is), and the whole  
thing is running perfectly as far as the traffic that runs through the  
cluster is not too high (that is, the number of packets/messages sent  
by the primary firewall equals the number of packets/messages received  
by the backup one).

Problems appears when I'm running load-heavy benchmarks on the primary  
firewall ; it generates much more traffic between the firewalls (in  
order to propagate state changes), and there is some packets/messages  
lost in the process.

I don't know the exact cause of this, but the problem doesn't appear  
when using UDP or TCP sockets instead of TIPC. No problem is showing  
in the system log either.
I thought maybe it could come from the either the sender or receiver  
buffer queue being congested, but then how can I verify this ? I've  
read from the docs that TIPC provides some debug functions, but then  
you have to recompile the module and I'm not sure of whether or not it  
would give me relevant informations ?

Thanks for reading me !

Regards,

Quentin Aebischer
University of Sherbrooke.
Canada