[Bloat] [Codel] [PATCH v12] codel: Controlled Delay AQM

[Eric Dumazet]

On Wed, 2012-05-09 at 08:47 -0700, Dave Taht wrote:
> Doing (or leveraging) the timestamp on actual ingress to the system,
> rather than entrance to the exiting queue, more accurately will
> compensate for overall system loading and the overhead of traversing
> 3+ pages of function calls to get it from the
> entrance point to the exit, and should result in 'tighter' results.


Most qdiscs dont need this timestamp, so it would add overhead in fast
path.

Also, the delta between 'entering' qdisc layer and entering codel
enqueue is almost a constant for a given qdisc/filter setup, and less
than one us :

If you really want to take it into account, replace 'target 5000us' by
'target 5001 us'