[Xorp-hackers] xorp-ospf performance issues: busy-spins & packet floods.

Pavlin Radoslavov pavlin at icir.org
Sat Oct 20 20:02:07 PDT 2007


Ben Greear <greearb at candelatech.com> wrote:

> Pavlin Radoslavov wrote:
> > See SelectorList::get_ready_priority() inside libxorp/selector.cc
> > for example.
> > It is used to find the best priority of a file descriptor that is
> > ready.
> >   
> 
> What do you think about having the run() method execute the timer(s),
> then tasks, then the selector.   I'd run the selector last since it can 
> sleep.
> We can get rid of the get_ready_priority() all together that way.  If you
> really wanted to not handle all of the available fds at once (ie, allow low
> priorities to be skipped), then we could just not process all of them.  
> I can't
> think of a good reason to do that, however.
> 
> Then, we have only one call to select(), as well as making sure that each
> of the run-queues (timers, tasks, selectors) gets some time to run.  I 
> am suspicious
> that the loop I see in ospf has to do with an endless timer loop where 
> the ospf
> code keeps trying to fire timers but never reads/writes the sockets.  I 
> also assume
> that if it could actually process the sockets, it's timers would be 
> satisfied and quit
> re-arming themselves.  This is idle conjecture at this time, but it 
> would fit the
> strace log I saw...

The problem should be approached by deciding first what kind of
semantics do we want from the eventloop.
Mark Handley will be a better person to query about the current set of
semantics (tasks, priority, etc), because he introduced and
implemented them so he can provide the insights and reasonings
behind various choices. 

Thanks,
Pavlin

> I think I can code this up and test it out if it's something you'd
> consider merging.
> 
> Thanks,
> Ben
> 
> -- 
> Ben Greear <greearb at candelatech.com> 
> Candela Technologies Inc  http://www.candelatech.com
> 
> 
> _______________________________________________
> Xorp-hackers mailing list
> Xorp-hackers at icir.org
> http://mailman.ICSI.Berkeley.EDU/mailman/listinfo/xorp-hackers



More information about the Xorp-hackers mailing list