Re: [PATCH] Speed-up pfifo_fast lookup using a bitmap

Re: [PATCH] Speed-up pfifo_fast lookup using a bitmap

[Krishna Kumar]

Hi Jarek,

Jarek Poplawski <jarkao2@gmail.com> wrote on 08/14/2009 04:31:27 PM:

> Alas, private or public, these values are lower on average than
> before, so I'm not sure the complexity (especially in reading) added
> by this patch is worth it. So, I can only say it looks formally OK,
> except the changelog and maybe 2 cosmetical suggestions below.

Maybe the different test parameters result in smaller improvements.
I agree with you - the first approach is very readable and probably
preferable, while the second introduces a new structure and more
complexity.

I am sending both these versions a last time, after fixing your
comments in case someone can help decide if one or the other is
better (sorry I forgot to run checkpatch couple of times, will try
to remember next time).

Thanks,

- KK

[PATCH v1] Speed-up pfifo_fast lookup using a public bitmap

[Krishna Kumar]

Signed-off-by: Krishna Kumar <krkumar2@in.ibm.com>
---

 include/net/sch_generic.h |    1 
 net/sched/sch_generic.c   |   46 +++++++++++++++++++++++-------------
 2 files changed, 31 insertions(+), 16 deletions(-)

diff -ruNp org/include/net/sch_generic.h new/include/net/sch_generic.h
--- org/include/net/sch_generic.h	2009-08-07 12:05:43.000000000 +0530
+++ new/include/net/sch_generic.h	2009-08-13 16:55:32.000000000 +0530
@@ -72,6 +72,7 @@ struct Qdisc
 	 * For performance sake on SMP, we put highly modified fields at the end
 	 */
 	unsigned long		state;
+	u32			bitmap;
 	struct sk_buff_head	q;
 	struct gnet_stats_basic bstats;
 	struct gnet_stats_queue	qstats;
diff -ruNp org/net/sched/sch_generic.c new/net/sched/sch_generic.c
--- org/net/sched/sch_generic.c	2009-08-07 12:05:43.000000000 +0530
+++ new/net/sched/sch_generic.c	2009-08-14 18:23:01.000000000 +0530
@@ -406,18 +406,28 @@ static const u8 prio2band[TC_PRIO_MAX+1]
 
 #define PFIFO_FAST_BANDS 3
 
-static inline struct sk_buff_head *prio2list(struct sk_buff *skb,
-					     struct Qdisc *qdisc)
+/*
+ * Convert a bitmap to the first band number where an skb is queued, where:
+ * 	bitmap=0 means there are no skbs on any band.
+ * 	bitmap=1 means there is an skb on band 0.
+ *	bitmap=7 means there are skbs on all 3 bands, etc.
+ */
+static const int bitmap2band[] = {-1, 0, 1, 0, 2, 0, 1, 0};
+
+static inline struct sk_buff_head *band2list(struct Qdisc *qdisc, int band)
 {
 	struct sk_buff_head *list = qdisc_priv(qdisc);
-	return list + prio2band[skb->priority & TC_PRIO_MAX];
+
+	return list + band;
 }
 
 static int pfifo_fast_enqueue(struct sk_buff *skb, struct Qdisc* qdisc)
 {
-	struct sk_buff_head *list = prio2list(skb, qdisc);
+	int band = prio2band[skb->priority & TC_PRIO_MAX];
+	struct sk_buff_head *list = band2list(qdisc, band);
 
 	if (skb_queue_len(list) < qdisc_dev(qdisc)->tx_queue_len) {
+		qdisc->bitmap |= (1 << band);
 		qdisc->q.qlen++;
 		return __qdisc_enqueue_tail(skb, qdisc, list);
 	}
@@ -427,14 +437,17 @@ static int pfifo_fast_enqueue(struct sk_
 
 static struct sk_buff *pfifo_fast_dequeue(struct Qdisc* qdisc)
 {
-	int prio;
-	struct sk_buff_head *list = qdisc_priv(qdisc);
+	int band = bitmap2band[qdisc->bitmap];
 
-	for (prio = 0; prio < PFIFO_FAST_BANDS; prio++) {
-		if (!skb_queue_empty(list + prio)) {
-			qdisc->q.qlen--;
-			return __qdisc_dequeue_head(qdisc, list + prio);
-		}
+	if (likely(band >= 0)) {
+		struct sk_buff_head *list = qdisc_priv(qdisc);
+		struct sk_buff *skb = __qdisc_dequeue_head(qdisc, list + band);
+
+		qdisc->q.qlen--;
+		if (skb_queue_empty(list + band))
+			qdisc->bitmap &= ~(1 << band);
+
+		return skb;
 	}
 
 	return NULL;
@@ -442,12 +455,12 @@ static struct sk_buff *pfifo_fast_dequeu
 
 static struct sk_buff *pfifo_fast_peek(struct Qdisc* qdisc)
 {
-	int prio;
-	struct sk_buff_head *list = qdisc_priv(qdisc);
+	int band = bitmap2band[qdisc->bitmap];
+
+	if (band >= 0) {
+		struct sk_buff_head *list = qdisc_priv(qdisc);
 
-	for (prio = 0; prio < PFIFO_FAST_BANDS; prio++) {
-		if (!skb_queue_empty(list + prio))
-			return skb_peek(list + prio);
+		return skb_peek(list + band);
 	}
 
 	return NULL;
@@ -461,6 +474,7 @@ static void pfifo_fast_reset(struct Qdis
 	for (prio = 0; prio < PFIFO_FAST_BANDS; prio++)
 		__qdisc_reset_queue(qdisc, list + prio);
 
+	qdisc->bitmap = 0;
 	qdisc->qstats.backlog = 0;
 	qdisc->q.qlen = 0;
 }

[PATCH v2] Speed-up pfifo_fast lookup using a private bitmap

[Krishna Kumar]

Signed-off-by: Krishna Kumar <krkumar2@in.ibm.com>
---

 net/sched/sch_generic.c |   70 ++++++++++++++++++++++++++------------
 1 file changed, 48 insertions(+), 22 deletions(-)

diff -ruNp org/net/sched/sch_generic.c new/net/sched/sch_generic.c
--- org/net/sched/sch_generic.c	2009-08-07 12:05:43.000000000 +0530
+++ new/net/sched/sch_generic.c	2009-08-14 18:23:16.000000000 +0530
@@ -406,18 +406,38 @@ static const u8 prio2band[TC_PRIO_MAX+1]
 
 #define PFIFO_FAST_BANDS 3
 
-static inline struct sk_buff_head *prio2list(struct sk_buff *skb,
-					     struct Qdisc *qdisc)
+/*
+ * Private data for a pfifo_fast scheduler containing:
+ * 	- the three band queues
+ * 	- bitmap indicating which of the bands contain skbs.
+ */
+struct pfifo_fast_priv {
+	u32 bitmap;
+	struct sk_buff_head q[PFIFO_FAST_BANDS];
+};
+
+/*
+ * Convert a bitmap to the first band number where an skb is queued, where:
+ * 	bitmap=0 means there are no skbs on any band.
+ * 	bitmap=1 means there is an skb on band 0.
+ *	bitmap=7 means there are skbs on all 3 bands, etc.
+ */
+static const int bitmap2band[] = {-1, 0, 1, 0, 2, 0, 1, 0};
+
+static inline struct sk_buff_head *band2list(struct pfifo_fast_priv *priv,
+					     int band)
 {
-	struct sk_buff_head *list = qdisc_priv(qdisc);
-	return list + prio2band[skb->priority & TC_PRIO_MAX];
+	return priv->q + band;
 }
 
 static int pfifo_fast_enqueue(struct sk_buff *skb, struct Qdisc* qdisc)
 {
-	struct sk_buff_head *list = prio2list(skb, qdisc);
+	int band = prio2band[skb->priority & TC_PRIO_MAX];
+	struct pfifo_fast_priv *priv = qdisc_priv(qdisc);
+	struct sk_buff_head *list = band2list(priv, band);
 
 	if (skb_queue_len(list) < qdisc_dev(qdisc)->tx_queue_len) {
+		priv->bitmap |= (1 << band);
 		qdisc->q.qlen++;
 		return __qdisc_enqueue_tail(skb, qdisc, list);
 	}
@@ -427,14 +447,18 @@ static int pfifo_fast_enqueue(struct sk_
 
 static struct sk_buff *pfifo_fast_dequeue(struct Qdisc* qdisc)
 {
-	int prio;
-	struct sk_buff_head *list = qdisc_priv(qdisc);
+	struct pfifo_fast_priv *priv = qdisc_priv(qdisc);
+	int band = bitmap2band[priv->bitmap];
 
-	for (prio = 0; prio < PFIFO_FAST_BANDS; prio++) {
-		if (!skb_queue_empty(list + prio)) {
-			qdisc->q.qlen--;
-			return __qdisc_dequeue_head(qdisc, list + prio);
-		}
+	if (likely(band >= 0)) {
+		struct sk_buff_head *list = band2list(priv, band);
+		struct sk_buff *skb = __qdisc_dequeue_head(qdisc, list);
+
+		qdisc->q.qlen--;
+		if (skb_queue_empty(list))
+			priv->bitmap &= ~(1 << band);
+
+		return skb;
 	}
 
 	return NULL;
@@ -442,12 +466,13 @@ static struct sk_buff *pfifo_fast_dequeu
 
 static struct sk_buff *pfifo_fast_peek(struct Qdisc* qdisc)
 {
-	int prio;
-	struct sk_buff_head *list = qdisc_priv(qdisc);
+	struct pfifo_fast_priv *priv = qdisc_priv(qdisc);
+	int band = bitmap2band[priv->bitmap];
+
+	if (band >= 0) {
+		struct sk_buff_head *list = band2list(priv, band);
 
-	for (prio = 0; prio < PFIFO_FAST_BANDS; prio++) {
-		if (!skb_queue_empty(list + prio))
-			return skb_peek(list + prio);
+		return skb_peek(list);
 	}
 
 	return NULL;
@@ -456,11 +481,12 @@ static struct sk_buff *pfifo_fast_peek(s
 static void pfifo_fast_reset(struct Qdisc* qdisc)
 {
 	int prio;
-	struct sk_buff_head *list = qdisc_priv(qdisc);
+	struct pfifo_fast_priv *priv = qdisc_priv(qdisc);
 
 	for (prio = 0; prio < PFIFO_FAST_BANDS; prio++)
-		__qdisc_reset_queue(qdisc, list + prio);
+		__qdisc_reset_queue(qdisc, band2list(priv, prio));
 
+	priv->bitmap = 0;
 	qdisc->qstats.backlog = 0;
 	qdisc->q.qlen = 0;
 }
@@ -480,17 +506,17 @@ nla_put_failure:
 static int pfifo_fast_init(struct Qdisc *qdisc, struct nlattr *opt)
 {
 	int prio;
-	struct sk_buff_head *list = qdisc_priv(qdisc);
+	struct pfifo_fast_priv *priv = qdisc_priv(qdisc);
 
 	for (prio = 0; prio < PFIFO_FAST_BANDS; prio++)
-		skb_queue_head_init(list + prio);
+		skb_queue_head_init(band2list(priv, prio));
 
 	return 0;
 }
 
 static struct Qdisc_ops pfifo_fast_ops __read_mostly = {
 	.id		=	"pfifo_fast",
-	.priv_size	=	PFIFO_FAST_BANDS * sizeof(struct sk_buff_head),
+	.priv_size	=	sizeof(struct pfifo_fast_priv),
 	.enqueue	=	pfifo_fast_enqueue,
 	.dequeue	=	pfifo_fast_dequeue,
 	.peek		=	pfifo_fast_peek,

Re: [PATCH] Speed-up pfifo_fast lookup using a bitmap

[Jarek Poplawski]

On Fri, Aug 14, 2009 at 06:54:58PM +0530, Krishna Kumar wrote:
> ďťżHi Jarek,
> 
> Jarek Poplawski <jarkao2@gmail.com> wrote on 08/14/2009 04:31:27 PM:
> 
> > Alas, private or public, these values are lower on average than
> > before, so I'm not sure the complexity (especially in reading) added
> > by this patch is worth it. So, I can only say it looks formally OK,
> > except the changelog and maybe 2 cosmetical suggestions below.
> 
> Maybe the different test parameters result in smaller improvements.
> I agree with you - the first approach is very readable and probably
> preferable, while the second introduces a new structure and more
> complexity.

Actually, I meant the complexity added by any of these versions. If
you declare: "This helps in faster lookup for a skb when there are no
high priority skbs.", there is a question if less than 1% is enough.

> I am sending both these versions a last time, after fixing your
> comments in case someone can help decide if one or the other is
> better (sorry I forgot to run checkpatch couple of times, will try
> to remember next time).

I definitely can't see a reason to make this variable public, and
prefer the private (v2) version (which still lacks a changelog, btw).

Thanks,
Jarek P.

Re: [PATCH] Speed-up pfifo_fast lookup using a bitmap

[David Miller]

From: Jarek Poplawski <jarkao2@gmail.com>
Date: Fri, 14 Aug 2009 23:36:43 +0200

> I definitely can't see a reason to make this variable public, and
> prefer the private (v2) version (which still lacks a changelog, btw).

I'm fine with patch v2, but yes it needs to be properly submitted
with a real changelog before I can apply it :-)

Re: [PATCH] Speed-up pfifo_fast lookup using a bitmap

[Krishna Kumar2]

Sorry about the delay, I am sick with the flu for the last few days.
I will send the patch tomorrow.

thanks,

- KK

David Miller <davem@davemloft.net> wrote on 08/18/2009 07:33:43 AM:

> David Miller <davem@davemloft.net>
> 08/18/2009 07:33 AM
>
> To
>
> jarkao2@gmail.com
>
> cc
>
> Krishna Kumar2/India/IBM@IBMIN, kaber@trash.net, netdev@vger.kernel.org,
> herbert@gondor.apana.org.au
>
> Subject
>
> Re: [PATCH] Speed-up pfifo_fast lookup using a bitmap
>
> From: Jarek Poplawski <jarkao2@gmail.com>
> Date: Fri, 14 Aug 2009 23:36:43 +0200
>
> > I definitely can't see a reason to make this variable public, and
> > prefer the private (v2) version (which still lacks a changelog, btw).
>
> I'm fine with patch v2, but yes it needs to be properly submitted
> with a real changelog before I can apply it :-)