On Tue, 2015-11-03 at 23:04 +0200, Andrew wrote: > Hi. > > This is common trouble due to hierarchical shapers realization (global > tree lock on packet dequeuing - so when one CPU looks for parent class > where tokens can be borrowed, other CPUs are waiting). It's mentioned > even in academic publications :) You can read about it here: > http://www.ijcset.com/docs/IJCSET13-04-04-113.pdf > > I think that simple lock removing will greatly improve performance; and > race conditions on packets dequeuing shouldn't hurt anything except > shaping accuracy. Another solutions looks more complex.
Thanks Andrew, I am very well aware of qdisc spinlock contention. And race conditions on packets dequeueing will _crash_ your host. Plainly. This is why I am advising using GRO in the first place. -- To unsubscribe from this list: send the line "unsubscribe netdev" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
