From mboxrd@z Thu Jan 1 00:00:00 1970 From: Patrick McHardy Subject: Re: [PATCH net-2.6.25] qdisc: new rate limiter Date: Sat, 08 Dec 2007 03:57:08 +0100 Message-ID: <475A0804.1060006@trash.net> References: <20071207162618.35715892@freepuppy.rosehill> Mime-Version: 1.0 Content-Type: text/plain; charset=ISO-8859-1; format=flowed Content-Transfer-Encoding: 7bit Cc: "David S. Miller" , netdev@vger.kernel.org To: Stephen Hemminger Return-path: Received: from stinky.trash.net ([213.144.137.162]:47077 "EHLO stinky.trash.net" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751006AbXLHC5o (ORCPT ); Fri, 7 Dec 2007 21:57:44 -0500 In-Reply-To: <20071207162618.35715892@freepuppy.rosehill> Sender: netdev-owner@vger.kernel.org List-ID: Stephen Hemminger wrote: > This is a time based rate limiter for use in network testing. When doing > network tests it is often useful to test at reduced bandwidths. The existing > Token Bucket Filter provides rate control, but causes bursty traffic that > can cause different performance than real world. Another alternative is > the PSPacer, but it depends on pause frames which may also cause issues. > > The qdisc depends on high resolution timers and clocks, so it will probably > use more CPU than others making it a poor choice for use when doing traffic > shaping for QOS. > > Signed-off-by: Stephen Hemminger > > --- a/include/linux/pkt_sched.h 2007-10-30 09:18:29.000000000 -0700 > +++ b/include/linux/pkt_sched.h 2007-12-07 13:37:50.000000000 -0800 > @@ -475,4 +475,10 @@ struct tc_netem_corrupt > > #define NETEM_DIST_SCALE 8192 > > +struct tc_rlim_qopt > +{ > + __u32 limit; /* fifo limit (packets) */ > + __u32 rate; /* bits per sec */ > This seems a bit small, 512mbit is the maximum rate. > --- /dev/null 1970-01-01 00:00:00.000000000 +0000 > +++ b/net/sched/sch_rlim.c 2007-12-07 16:22:10.000000000 -0800 > @@ -0,0 +1,350 @@ > +static struct sk_buff *rlim_dequeue(struct Qdisc *sch) > +{ > + struct rlim_sched_data *q = qdisc_priv(sch); > + struct sk_buff *skb; > + ktime_t now = ktime_get(); > + > + /* if haven't reached the correct time slot, start timer */ > + if (now.tv64 < q->next_send.tv64) { > + sch->flags |= TCQ_F_THROTTLED; > + hrtimer_start(&q->watchdog.timer, q->next_send, > + HRTIMER_MODE_ABS); > + return NULL; > + } > + > + skb = q->qdisc->dequeue(q->qdisc); > + if (skb) { > + q->next_send = ktime_add_ns(now, pkt_time(q, skb)); > + sch->flags &= ~TCQ_F_THROTTLED; > qlen is not decremented here. > + } > + return skb; > +} > + > +static int rlim_requeue(struct sk_buff *skb, struct Qdisc *sch) > +{ > + struct rlim_sched_data *q = qdisc_priv(sch); > + int ret; > + > + ret = q->qdisc->ops->requeue(skb, q->qdisc); > + if (!ret) { > + q->next_send = ktime_sub_ns(q->next_send, pkt_time(q, skb)); > + sch->q.qlen++; > + sch->qstats.requeues++; > + } > + > + return ret; > +} > + > +static void rlim_reset(struct Qdisc *sch) > +{ > + struct rlim_sched_data *q = qdisc_priv(sch); > + > + qdisc_reset_queue(sch); This should reset the child. > + > + q->next_send = ktime_get(); > + qdisc_watchdog_cancel(&q->watchdog); > +} > + > +static int rlim_change(struct Qdisc *sch, struct rtattr *opt) > +{ > + struct rlim_sched_data *q = qdisc_priv(sch); > + const struct tc_rlim_qopt *qopt; > + int err; > + > + if (opt == NULL || RTA_PAYLOAD(opt) < sizeof(struct tc_rlim_qopt)) > + return -EINVAL; > + > + qopt = RTA_DATA(opt); > Using nested attributes would make sure we don't run into problems with extensibility. > + err = set_fifo_limit(q->qdisc, qopt->limit); > + if (err) > + return err; > + > + q->limit = qopt->limit; > + if (qopt->rate == 0) > + q->cost = 0; /* unlimited */ > + else { > + q->cost = (u64)NSEC_PER_SEC << NSEC_SCALE; > + do_div(q->cost, qopt->rate); > + } > + > + pr_debug("rlim_change: rate=%u cost=%llu\n", > + qopt->rate, q->cost); > + > + return 0; > +} > + > +static struct Qdisc_class_ops rlim_class_ops = { > This can be const. > + .graft = rlim_graft, > + .leaf = rlim_leaf, > + .get = rlim_get, > + .put = rlim_put, > + .change = rlim_change_class, > + .delete = rlim_delete, > + .walk = rlim_walk, > + .tcf_chain = rlim_find_tcf, > + .dump = rlim_dump_class, > +}; > > >