Re: [RFC] sched: CHOKe packet scheduler (v0.6)

Le mercredi 12 janvier 2011 à 09:27 -0800, Stephen Hemminger a écrit :

On Wed, 12 Jan 2011 08:13:48 +0100
Eric Dumazet [off-list ref] wrote:

quoted

Hi Stephen, here is my v0.6 version :

- Added sanity checks before kcalloc()/kzalloc()
- Added a __GFP_NOWARN to kcalloc()
- Added call to qdisc_bstats_update() after commit bfe0d0298f2a67d94d5
(net_sched: factorize qdisc stats handling)

TODO :
- Added a stat specific update to track CHOKe probabilistic dual-drops
  I temporarily use requeues counter to make sure our code works

I am going to redo stats and config.
Should thresholds be packet or byte based? I prefer packet
Also leaning towards merging qmax and qlimit together.

I believe CHOKe spirit is per packet, I agree with you.

(This makes the tab[] array sizing directly depends on ctl->limit, not
on a computation ctl->limit/smallest_packet_size)

Not sure we can merge qmax and qlimit, are you sure its OK with CHOKe
paper and experimental results ?
(Sorry I cant spend too much time right now to check this point)