Re: Kernel rwlock design, Multicore and IGMP

[Eric Dumazet]

Le mercredi 17 novembre 2010 Ã 09:30 +0800, Cypher Wu a Ãcrit :
> O
> struct in_device::mr_ifc_timer. Every time when a process join/leave a
> MC group, igmp_ifc_event() -> igmp_ifc_start_timer() will start the
> timer on the core that the system call issued, and
> igmp_ifc_timer_expire() will be called on that core in the bottem halt
> of timer interrupt.
> If we call join/leave on mutlicores that timers will run on all these
> cores, but it seems only one or two will generate IGMP message, others
> will only lock the list and loop throught it with nothing generated.
> 
> 

Problem would not be timer being restarted on different cores (very
small impact), but scanning a list in igmpv3_send_cr() with many items
in it and expensive things, under timer handler (softirq), so adding
spikes of latency.

IGMP_Unsolicited_Report_Interval is 10 seconds, so we start timer in a 5
second average.

I am not sure there is a need to join/leave thousand of groups per
second anyway...



--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/