[PATCH] pkt_sched: fix bandwidth's accuracy of software device with htb

[PATCH] pkt_sched: fix bandwidth's accuracy of software device with htb

[Yang Yingliang]

When we set bandwidth on the software devices with htb,
it's not precise.

With htb, sch->limit is set from devices TX queue's
length that is not 0. Because software devices TX
queue is 0, sch->limit is set to default value 1.

Because the length of skb_buff queue is bigger than
sch->limit whose value is 1, some packets are dropped.
This leads to software devices can't make full use of
the bandwidth and influence the accuracy.

To fix this issue, set the default value of sch->limit
to 1000 that is the value of eth device's tx_queueu_len.

Example:
tc qdisc add dev bond0 root handle 1: htb default 1
tc class add dev bond0 parent 1:0 classid 1:1 htb rate 1Gbit ceil 1Gbit
iperf -c host -t 30 -i 10

With old sch->limit:
[ ID] Interval       Transfer     Bandwidth
[  3]  0.0-10.0 sec   579 MBytes   486 Mbits/sec
[  3] 10.0-20.0 sec   604 MBytes   507 Mbits/sec
[  3] 20.0-30.0 sec   551 MBytes   462 Mbits/sec
[  3]  0.0-30.0 sec  1.69 GBytes   485 Mbits/sec

With new sch->limit:
[ ID] Interval       Transfer     Bandwidth
[  3]  0.0-10.0 sec  1.11 GBytes   955 Mbits/sec
[  3] 10.0-20.0 sec  1.11 GBytes   955 Mbits/sec
[  3] 20.0-30.0 sec  1.11 GBytes   955 Mbits/sec
[  3]  0.0-30.0 sec  3.33 GBytes   955 Mbits/sec

Signed-off-by: Yang Yingliang <yangyingliang@huawei.com>
---
 net/sched/sch_fifo.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/net/sched/sch_fifo.c b/net/sched/sch_fifo.c
index e15a9eb..8dd27ec 100644
--- a/net/sched/sch_fifo.c
+++ b/net/sched/sch_fifo.c
@@ -54,7 +54,7 @@ static int fifo_init(struct Qdisc *sch, struct nlattr *opt)
 	bool is_bfifo = sch->ops == &bfifo_qdisc_ops;
 
 	if (opt == NULL) {
-		u32 limit = qdisc_dev(sch)->tx_queue_len ? : 1;
+		u32 limit = qdisc_dev(sch)->tx_queue_len ? : 1000;
 
 		if (is_bfifo)
 			limit *= psched_mtu(qdisc_dev(sch));
-- 
1.7.12



.

Re: [PATCH] pkt_sched: fix bandwidth's accuracy of software device with htb

[Eric Dumazet]

On Wed, 2013-07-31 at 10:39 +0800, Yang Yingliang wrote:
> When we set bandwidth on the software devices with htb,
> it's not precise.
> 
> With htb, sch->limit is set from devices TX queue's
> length that is not 0. Because software devices TX
> queue is 0, sch->limit is set to default value 1.
> 
> Because the length of skb_buff queue is bigger than
> sch->limit whose value is 1, some packets are dropped.
> This leads to software devices can't make full use of
> the bandwidth and influence the accuracy.
> 
> To fix this issue, set the default value of sch->limit
> to 1000 that is the value of eth device's tx_queueu_len.
> 
> Example:
> tc qdisc add dev bond0 root handle 1: htb default 1
> tc class add dev bond0 parent 1:0 classid 1:1 htb rate 1Gbit ceil 1Gbit
> iperf -c host -t 30 -i 10
> 
> With old sch->limit:
> [ ID] Interval       Transfer     Bandwidth
> [  3]  0.0-10.0 sec   579 MBytes   486 Mbits/sec
> [  3] 10.0-20.0 sec   604 MBytes   507 Mbits/sec
> [  3] 20.0-30.0 sec   551 MBytes   462 Mbits/sec
> [  3]  0.0-30.0 sec  1.69 GBytes   485 Mbits/sec
> 
> With new sch->limit:
> [ ID] Interval       Transfer     Bandwidth
> [  3]  0.0-10.0 sec  1.11 GBytes   955 Mbits/sec
> [  3] 10.0-20.0 sec  1.11 GBytes   955 Mbits/sec
> [  3] 20.0-30.0 sec  1.11 GBytes   955 Mbits/sec
> [  3]  0.0-30.0 sec  3.33 GBytes   955 Mbits/sec
> 
> Signed-off-by: Yang Yingliang <yangyingliang@huawei.com>
> ---
>  net/sched/sch_fifo.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/net/sched/sch_fifo.c b/net/sched/sch_fifo.c
> index e15a9eb..8dd27ec 100644
> --- a/net/sched/sch_fifo.c
> +++ b/net/sched/sch_fifo.c
> @@ -54,7 +54,7 @@ static int fifo_init(struct Qdisc *sch, struct nlattr *opt)
>  	bool is_bfifo = sch->ops == &bfifo_qdisc_ops;
>  
>  	if (opt == NULL) {
> -		u32 limit = qdisc_dev(sch)->tx_queue_len ? : 1;
> +		u32 limit = qdisc_dev(sch)->tx_queue_len ? : 1000;
>  
>  		if (is_bfifo)
>  			limit *= psched_mtu(qdisc_dev(sch));

Hmm. This looks like user error to me.

What if existing scripts relied on the limit being 1 ?

If you want to use pfifo on a virtual device, while not setting explicit
limit, you have to "ifconfig bond0 txqueuelen 1000"

Seriously, using pfifo or bfifo without setting explicit plimit/blimit
is not very safe, even with non virtual devices.