From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: X-Spam-Checker-Version: SpamAssassin 3.4.0 (2014-02-07) on aws-us-west-2-korg-lkml-1.web.codeaurora.org X-Spam-Level: X-Spam-Status: No, score=-6.8 required=3.0 tests=DKIM_SIGNED,DKIM_VALID, DKIM_VALID_AU,FREEMAIL_FORGED_FROMDOMAIN,FREEMAIL_FROM, HEADER_FROM_DIFFERENT_DOMAINS,INCLUDES_PATCH,MAILING_LIST_MULTI,SIGNED_OFF_BY, SPF_PASS autolearn=ham autolearn_force=no version=3.4.0 Received: from mail.kernel.org (mail.kernel.org [198.145.29.99]) by smtp.lore.kernel.org (Postfix) with ESMTP id 2742BC282CE for ; Mon, 8 Apr 2019 05:31:48 +0000 (UTC) Received: from vger.kernel.org (vger.kernel.org [209.132.180.67]) by mail.kernel.org (Postfix) with ESMTP id BE07020870 for ; Mon, 8 Apr 2019 05:31:47 +0000 (UTC) Authentication-Results: mail.kernel.org; dkim=pass (2048-bit key) header.d=gmail.com header.i=@gmail.com header.b="QJDFarBW" Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1726133AbfDHFbq (ORCPT ); Mon, 8 Apr 2019 01:31:46 -0400 Received: from mail-qk1-f195.google.com ([209.85.222.195]:33706 "EHLO mail-qk1-f195.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1725993AbfDHFbq (ORCPT ); Mon, 8 Apr 2019 01:31:46 -0400 Received: by mail-qk1-f195.google.com with SMTP id k189so7252126qkc.0 for ; Sun, 07 Apr 2019 22:31:45 -0700 (PDT) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gmail.com; s=20161025; h=mime-version:references:in-reply-to:from:date:message-id:subject:to :cc:content-transfer-encoding; bh=J3vD9QQEpHLYSf62HUv8Dr7ncPrQfaQyagDbNhmgwOY=; b=QJDFarBW+gcg9PuW/minkGetty/yjUWI/hG8V/J3VYPnk5qJ/8VAeeiTZRROROaz5n zkzF2FqZyqWHrA4hMDqz0lvdDRP7wE8jMf0W3XS5mFLKgdD3/w1D2ruBd94AKf4MjEbM 3+qCRYs5GvzKleoSegVsoYwswvE7QN/rOtLbDnSm9ZoqC8vVQ4MwPMs3CNJZuDoGx5iw zV1XrsYopl8XX53GeyKsrW5SGPa1plsXpsVbx4tKKjtApUeyv6plj3eRiVGpFo+0lm3I CVglq7iznT1Tgt+yFoXuJtG2p3lrNPk/KFF1t9uZavB5vaPdVYue9u1/j0E2D8MCXxxU 1BNQ== X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20161025; h=x-gm-message-state:mime-version:references:in-reply-to:from:date :message-id:subject:to:cc:content-transfer-encoding; bh=J3vD9QQEpHLYSf62HUv8Dr7ncPrQfaQyagDbNhmgwOY=; b=rOoN/Fkc7tVuKfUJ6s0nuADsznnK+fY0FI8z5VGfd+UaPS5kP9LXG8KmiOeh7HyKCV 5SussncJT/bs1z/G1NpkyRTj755KlTEJoZGcSaLePqxdLkk6FrGkY9NnKdjDSxG7eGdk JFcZrQ3/inXI0EdUskOTLMAHC0/JkzF1LhnQXN6slVDXvDY/IqQ4E5t1r6ytxTDZzuH7 Ia3/62COSSXzoxMU4q4AvKdhGSlYqSObjcQKhppMcMsNsa4vieG1hMToAQ33eXaosjyb u75LbfdL0MOMEADIrmezi8lKKN9eygWSTNEYYrJx+q5bd6E+WdSLVWTOjaffB8Tb0BlW V2zQ== X-Gm-Message-State: APjAAAVZUKQjt+Nel7ZzZ5rXqT/9t0/aDMwFAsft6RxrJCsAyPcVLY2o ZcSg/53GUKmOap2oSA6CglXG34KszUftj+i5JSk= X-Google-Smtp-Source: APXvYqxJ/wNeHiptJCqYgtlKcosjGWJ9l/ruxdKAP8Yh++VOjiPLL54xiuZjE/ukVnUyDpAwVMA7TTc+DwQIC9OrKKs= X-Received: by 2002:a05:620a:15f5:: with SMTP id p21mr20749397qkm.5.1554701504559; Sun, 07 Apr 2019 22:31:44 -0700 (PDT) MIME-Version: 1.0 References: <20190331174005.5841-1-gautamramk@gmail.com> <20190331174005.5841-3-gautamramk@gmail.com> <87bm1optqu.fsf@toke.dk> <87k1gcnwt6.fsf@toke.dk> In-Reply-To: <87k1gcnwt6.fsf@toke.dk> From: Gautam Ramakrishnan Date: Mon, 8 Apr 2019 11:01:32 +0530 Message-ID: Subject: Re: [RFC net-next 2/2] net: sched: fq_pie: Flow Queue PIE AQM To: =?UTF-8?B?VG9rZSBIw7hpbGFuZC1Kw7hyZ2Vuc2Vu?= Cc: Jamal Hadi Salim , davem@davemloft.net, netdev@vger.kernel.org, "Mohit P . Tahiliani" , "Sachin D . Patil" , Mohit Bhasi , "V . Saicharan" , Leslie Monis , Dave Taht Content-Type: text/plain; charset="UTF-8" Content-Transfer-Encoding: quoted-printable Sender: netdev-owner@vger.kernel.org Precedence: bulk List-ID: X-Mailing-List: netdev@vger.kernel.org I was trying to refactor the code and I ran into some issues. 1. I moved some of the parameters such as flows_cnt into a new struct called fq_pie_params, instead of keeping them in fq_pie_sched_data. Should I move those parameters back into fq_pie_sched_data? 2. fq_codel maintains the backlog variable as a list in the fq_codel_sched_data, whereas I maintain a backlog in the struct fq_pie_flow. What approach should I follow? 3. Would maintaining a per flow pie_stats be useful in the future? I do not use the per flow stats anywhere in the code. On Tue, Apr 2, 2019 at 10:55 PM Toke H=C3=B8iland-J=C3=B8rgensen wrote: > > Gautam Ramakrishnan writes: > > > Hello, thanks for the feedback > > > > On Tue, Apr 2, 2019 at 4:19 PM Toke H=C3=B8iland-J=C3=B8rgensen wrote: > >> > >> Some suggestions below to make fq_pie and fq_codel more similar (ref. = my > >> previous email). > >> > >> Also, a few unrelated nits. > >> > >> > From: Mohit P. Tahiliani > >> > > >> > FQ-PIE incorporates fair/flow queuing in which every queue > >> > is managed by an instance of PIE queueing discipline. > >> > The algorithm provides good control over the queueing delay > >> > while at the same time, ensures fairness across various > >> > flows sharing the same link. > >> > > >> > Principles: > >> > - Packets are classified stochastically on flows. > >> > - Each flow has a PIE managed queue. > >> > - Flows are linked onto two (Round Robin) lists, > >> > so that new flows have priority on old ones. > >> > - For a given flow, packets are dropped on arrival according > >> > to a drop probability. > >> > - Tail drops only. > >> > >> Why tail drops only? > > > > I had mentioned this because packets are dropped only at enqueue. > > Yup, realise that; was just wondering why you went with this design > instead of doing the head drop that fq_codel does? > > >> > >> > Usage: > >> > tc qdisc ... fq_pie [ limit PACKETS ] [ flows NUMBER ] > >> > [ alpha NUMBER ] [ beta NUMBER ] > >> > [ target TIME us ] [ tupdate TIME us ] > >> > [ bytemode ] [ quantum BYTES ] > >> > [ ecn | ecn_prob PERCENTAGE ] > >> > > >> > defaults: 1024 flows, 10240 packets limit, quantum : device MTU > >> > target: 15ms > >> > tupdate: 15ms > >> > alpha: 2 (on a scale of 0 to 16) > >> > beta: 20 (on a scale of 0 to 16) > >> > ecn: false > >> > ecn_prob: 10% > >> > > >> > Signed-off-by: Mohit P. Tahiliani > >> > Signed-off-by: Sachin D. Patil > >> > Signed-off-by: Mohit Bhasi > >> > Signed-off-by: V. Saicharan > >> > Signed-off-by: Leslie Monis > >> > Signed-off-by: Gautam Ramakrishnan > >> > Cc: Dave Taht > >> > --- > >> > include/uapi/linux/pkt_sched.h | 28 ++ > >> > net/sched/Kconfig | 14 +- > >> > net/sched/Makefile | 1 + > >> > net/sched/sch_fq_pie.c | 485 ++++++++++++++++++++++++++++++= +++ > >> > 4 files changed, 527 insertions(+), 1 deletion(-) > >> > create mode 100644 net/sched/sch_fq_pie.c > >> > > >> > diff --git a/include/uapi/linux/pkt_sched.h b/include/uapi/linux/pkt= _sched.h > >> > index 7ee74c3474bf..005413bd09ee 100644 > >> > --- a/include/uapi/linux/pkt_sched.h > >> > +++ b/include/uapi/linux/pkt_sched.h > >> > @@ -964,6 +964,34 @@ struct tc_pie_xstats { > >> > __u32 ecn_mark; /* packets marked with ecn*/ > >> > }; > >> > > >> > +/* FQ PIE */ > >> > +enum { > >> > + TCA_FQ_PIE_UNSPEC, > >> > + TCA_FQ_PIE_TARGET, > >> > + TCA_FQ_PIE_LIMIT, > >> > + TCA_FQ_PIE_TUPDATE, > >> > + TCA_FQ_PIE_ALPHA, > >> > + TCA_FQ_PIE_BETA, > >> > + TCA_FQ_PIE_ECN, > >> > + TCA_FQ_PIE_QUANTUM, > >> > + TCA_FQ_PIE_BYTEMODE, > >> > + TCA_FQ_PIE_FLOWS, > >> > + TCA_FQ_PIE_ECN_PROB, > >> > + __TCA_FQ_PIE_MAX > >> > +}; > >> > +#define TCA_FQ_PIE_MAX (__TCA_FQ_PIE_MAX - 1) > >> > + > >> > +struct tc_fq_pie_xstats { > >> > + __u32 packets_in; /* total number of packets enqueued */ > >> > + __u32 dropped; /* packets dropped due to fq_pie_actio= n */ > >> > + __u32 overlimit; /* dropped due to lack of space in que= ue */ > >> > + __u32 ecn_mark; /* packets marked with ecn*/ > >> > + __u32 new_flow_count; /* number of time packets > >> > + created a 'new flow' */ > >> > + __u32 new_flows_len; /* count of flows in new list */ > >> > + __u32 old_flows_len; /* count of flows in old list */ > >> > +}; > >> > + > >> > /* CBS */ > >> > struct tc_cbs_qopt { > >> > __u8 offload; > >> > diff --git a/net/sched/Kconfig b/net/sched/Kconfig > >> > index 5c02ad97ef23..49f4dd9894a0 100644 > >> > --- a/net/sched/Kconfig > >> > +++ b/net/sched/Kconfig > >> > @@ -358,13 +358,25 @@ config NET_SCH_PIE > >> > help > >> > Say Y here if you want to use the Proportional Integral cont= roller > >> > Enhanced scheduler packet scheduling algorithm. > >> > - For more information, please see https://tools.ietf.org/html= /rfc8033 > >> > + For more information, please see > >> > + http://tools.ietf.org/html/draft-pan-tsvwg-pie-00 > >> > > >> > To compile this driver as a module, choose M here: the modul= e > >> > will be called sch_pie. > >> > > >> > If unsure, say N. > >> > > >> > +config NET_SCH_FQ_PIE > >> > + tristate "Flow Queue Proportional Integral controller Enhanced= (FQ-PIE) scheduler" > >> > + help > >> > + Say Y here if you want to use the Flow Queue Proportional In= tegral controller > >> > + Enhanced scheduler packet scheduling algorithm. > >> > + > >> > + To compile this driver as a module, choose M here: the modul= e > >> > + will be called sch_fq_pie. > >> > + > >> > + If unsure, say N. > >> > + > >> > config NET_SCH_INGRESS > >> > tristate "Ingress/classifier-action Qdisc" > >> > depends on NET_CLS_ACT > >> > diff --git a/net/sched/Makefile b/net/sched/Makefile > >> > index 8a40431d7b5c..fdcd3f7b2fb2 100644 > >> > --- a/net/sched/Makefile > >> > +++ b/net/sched/Makefile > >> > @@ -55,6 +55,7 @@ obj-$(CONFIG_NET_SCH_CAKE) +=3D sch_cake.o > >> > obj-$(CONFIG_NET_SCH_FQ) +=3D sch_fq.o > >> > obj-$(CONFIG_NET_SCH_HHF) +=3D sch_hhf.o > >> > obj-$(CONFIG_NET_SCH_PIE) +=3D sch_pie.o > >> > +obj-$(CONFIG_NET_SCH_FQ_PIE) +=3D sch_fq_pie.o > >> > obj-$(CONFIG_NET_SCH_CBS) +=3D sch_cbs.o > >> > obj-$(CONFIG_NET_SCH_ETF) +=3D sch_etf.o > >> > obj-$(CONFIG_NET_SCH_TAPRIO) +=3D sch_taprio.o > >> > diff --git a/net/sched/sch_fq_pie.c b/net/sched/sch_fq_pie.c > >> > new file mode 100644 > >> > index 000000000000..4ccefa0bc7f0 > >> > --- /dev/null > >> > +++ b/net/sched/sch_fq_pie.c > >> > @@ -0,0 +1,485 @@ > >> > +/* > >> > + * net/sched/sch_fq_pie.c > >> > + * > >> > + * This program is free software; you can redistribute it and/or > >> > + * modify it under the terms of the GNU General Public License > >> > + * as published by the Free Software Foundation; either version 2 > >> > + * of the License. > >> > + * > >> > + * This program is distributed in the hope that it will be useful, > >> > + * but WITHOUT ANY WARRANTY; without even the implied warranty of > >> > + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the > >> > + * GNU General Public License for more details. > >> > >> Lose the license boilerplate and replace it with a SPDX header line. > > > > I shall do that. > > > >> > >> > + * Author: Mohit P. Tahiliani > >> > + * Author: Sachin D. Patil > >> > + * Author: Mohit Bhasi > >> > + * Author: V Saicharan > >> > + * Author: Leslie Monis > >> > + * Author: Gautam Ramakrishnan > >> > + * > >> > + * References: > >> > + * RFC 8033: https://tools.ietf.org/html/rfc8033 > >> > + */ > >> > + > >> > +#include > >> > +#include > >> > +#include > >> > + > >> > +struct fq_pie_params { > >> > + struct pie_params p_params; > >> > + u32 ecn_prob; > >> > + u32 flows_cnt; > >> > +}; > >> > + > >> > +struct fq_pie_stats { > >> > + u32 packets_in; /* total number of packets enqueued */ > >> > + u32 dropped; /* packets dropped due to fq_pie actio= n */ > >> > + u32 overlimit; /* dropped due to lack of space in que= ue */ > >> > + u32 ecn_mark; /* packets marked with ECN */ > >> > + u32 new_flow_count; /* number of time packets created a ne= w flow */ > >> > +}; > >> > + > >> > +struct fq_pie_flow { > >> > + s32 deficit; /* number of credits remaining for the= flow */ > >> > + u32 backlog; /* size of data in the flow */ > >> > + u32 qlen; /* number of packets in the flow */ > >> > + struct sk_buff *head; > >> > + struct sk_buff *tail; > >> > + struct list_head flowchain; > >> > + struct pie_vars vars; /* pie vars for the flow */ > >> > + struct pie_stats stats; /* pie stats for the flow */ > >> > +}; > >> > + > >> > +struct fq_pie_sched_data { > >> > + u32 quantum; /* number of credits in deficit round = robin */ > >> > + struct fq_pie_flow *flows; > >> > + struct Qdisc *sch; > >> > + struct fq_pie_params params; > >> > + struct fq_pie_stats stats; > >> > + struct list_head old_flows; > >> > + struct list_head new_flows; > >> > + struct timer_list adapt_timer; > >> > +}; > >> > >> The flow and sched_data structs have quite a bit in common with those = in > >> fq_codel; but the members are in a different order. > >> > >> > +static void fq_pie_params_init(struct fq_pie_params *params) > >> > +{ > >> > + pie_params_init(¶ms->p_params); > >> > + params->ecn_prob =3D 10; > >> > + params->flows_cnt =3D 1024; > >> > +} > >> > + > >> > +static inline void flow_queue_add(struct fq_pie_flow *flow, > >> > + struct sk_buff *skb) > >> > +{ > >> > + if (!flow->head) > >> > + flow->head =3D skb; > >> > + else > >> > + flow->tail->next =3D skb; > >> > + flow->tail =3D skb; > >> > + skb->next =3D NULL; > >> > +} > >> > + > >> > +static int fq_pie_qdisc_enqueue(struct sk_buff *skb, struct Qdisc *= sch, > >> > + struct sk_buff **to_free) > >> > +{ > >> > + struct fq_pie_sched_data *q =3D qdisc_priv(sch); > >> > + struct fq_pie_flow *sel_flow; > >> > + u32 pkt_len; > >> > + u32 idx; > >> > + u8 enqueue =3D false; > >> > + > >> > + /* Classifies packet into corresponding flow */ > >> > + idx =3D reciprocal_scale(skb_get_hash(skb), q->params.flows_cn= t); > >> > + sel_flow =3D &q->flows[idx]; > >> > >> This is missing the ability to override the classification from tc > >> filters. See fq_codel_classify(). > > > > I wanted to keep it simple initially. Shall add that. > > > >> > >> > + > >> > + /* Checks if the qdisc is full */ > >> > + if (unlikely(qdisc_qlen(sch) >=3D sch->limit)) { > >> > + q->stats.overlimit++; > >> > + sel_flow->stats.overlimit++; > >> > + goto out; > >> > + } > >> > >> The memory_limit checks in fq_codel have turned out to be quite useful > >> on constrained systems. I'd suggest adding them here as well. > > > > I shall add that too. > > > >> > >> > + > >> > + if (!drop_early(sch, sel_flow->backlog, skb->len, &sel_flow->v= ars, > >> > + &q->params.p_params)) { > >> > + enqueue =3D true; > >> > + } else if (q->params.p_params.ecn && > >> > + sel_flow->vars.prob <=3D > >> > + (MAX_PROB / 100) * q->params.ecn_prob && > >> > + INET_ECN_set_ce(skb)) { > >> > + /* If packet is ecn capable, mark it if drop probabili= ty > >> > + * is lower than the parameter ecn_prob, else drop it. > >> > + */ > >> > + q->stats.ecn_mark++; > >> > + sel_flow->stats.ecn_mark++; > >> > + enqueue =3D true; > >> > + } > >> > + if (enqueue) { > >> > + pkt_len =3D qdisc_pkt_len(skb); > >> > + q->stats.packets_in++; > >> > + sch->qstats.backlog +=3D pkt_len; > >> > + sch->q.qlen++; > >> > + flow_queue_add(sel_flow, skb); > >> > + if (list_empty(&sel_flow->flowchain)) { > >> > + list_add_tail(&sel_flow->flowchain, &q->new_fl= ows); > >> > + q->stats.new_flow_count++; > >> > + sel_flow->deficit =3D q->quantum; > >> > + sel_flow->stats.dropped =3D 0; > >> > + sel_flow->qlen =3D 0; > >> > + sel_flow->backlog =3D 0; > >> > + } > >> > + sel_flow->qlen++; > >> > + sel_flow->stats.packets_in++; > >> > + sel_flow->backlog +=3D pkt_len; > >> > + return NET_XMIT_SUCCESS; > >> > + } > >> > +out: > >> > + q->stats.dropped++; > >> > + sel_flow->stats.dropped++; > >> > + return qdisc_drop(skb, sch, to_free); > >> > >> You probably want to return NET_XMIT_CN instead of NET_XMIT_DROP here. > >> > > > > Should I replace qdisc_drop with __qdisc_drop and return NET_XMIT_CN? > > Yeah, I think that would be more appropriate here, as that would > directly throttle sockets on the same host. > > >> > +} > >> > + > >> > +static const struct nla_policy fq_pie_policy[TCA_FQ_PIE_MAX + 1] = =3D { > >> > + [TCA_FQ_PIE_TARGET] =3D {.type =3D NLA_U32}, > >> > + [TCA_FQ_PIE_LIMIT] =3D {.type =3D NLA_U32}, > >> > + [TCA_FQ_PIE_TUPDATE] =3D {.type =3D NLA_U32}, > >> > + [TCA_FQ_PIE_ALPHA] =3D {.type =3D NLA_U32}, > >> > + [TCA_FQ_PIE_BETA] =3D {.type =3D NLA_U32}, > >> > + [TCA_FQ_PIE_ECN] =3D {.type =3D NLA_U32}, > >> > + [TCA_FQ_PIE_QUANTUM] =3D {.type =3D NLA_U32}, > >> > + [TCA_FQ_PIE_BYTEMODE] =3D {.type =3D NLA_U32}, > >> > + [TCA_FQ_PIE_FLOWS] =3D {.type =3D NLA_U32}, > >> > + [TCA_FQ_PIE_ECN_PROB] =3D {.type =3D NLA_U32} > >> > +}; > >> > + > >> > +static inline struct sk_buff *dequeue_head(struct fq_pie_flow *flow= ) > >> > +{ > >> > + struct sk_buff *skb =3D flow->head; > >> > + > >> > + flow->head =3D skb->next; > >> > + skb->next =3D NULL; > >> > + return skb; > >> > +} > >> > + > >> > +static struct sk_buff *fq_pie_qdisc_dequeue(struct Qdisc *sch) > >> > +{ > >> > + struct fq_pie_sched_data *q =3D qdisc_priv(sch); > >> > + struct sk_buff *skb =3D NULL; > >> > + struct fq_pie_flow *flow; > >> > + struct list_head *head; > >> > + u32 pkt_len; > >> > + > >> > +begin: > >> > + head =3D &q->new_flows; > >> > + if (list_empty(head)) { > >> > + head =3D &q->old_flows; > >> > + if (list_empty(head)) > >> > + return NULL; > >> > + } > >> > + > >> > + flow =3D list_first_entry(head, struct fq_pie_flow, flowchain)= ; > >> > + /* Flow has exhausted all its credits */ > >> > + if (flow->deficit <=3D 0) { > >> > + flow->deficit +=3D q->quantum; > >> > + list_move_tail(&flow->flowchain, &q->old_flows); > >> > + goto begin; > >> > + } > >> > + > >> > + if (flow->head) { > >> > + skb =3D dequeue_head(flow); > >> > + pkt_len =3D qdisc_pkt_len(skb); > >> > + sch->qstats.backlog -=3D pkt_len; > >> > + sch->q.qlen--; > >> > + qdisc_bstats_update(sch, skb); > >> > + } > >> > >> If you factor this out into a dequeue_func(), this dequeue() function = is > >> very close to identical to the one in fq_codel(). > > > > Do you suggest I put the if(flow->head) block in a different function? > > Yeah, that was what I meant. However, without doing the full refactoring > it's only borderline useful, so feel free to just keep it the way it > is... > > -Toke --=20 ------------- Gautam |