[PATCH 3/3] tcp_cubic: fix low utilization of CUBIC with HyStart

[PATCH 3/3] tcp_cubic: fix low utilization of CUBIC with HyStart

[Sangtae Ha]

HyStart sets the initial exit point of slow start.
Suppose that HyStart exits at 0.5BDP in a BDP network and no history exists.
If the BDP of a network is large, CUBIC's initial cwnd growth may be
too conservative to utilize the link.
CUBIC increases the cwnd 20% per RTT in this case.

Signed-off-by: Sangtae Ha <sangtae.ha@gmail.com>
---
 net/ipv4/tcp_cubic.c |    9 +++++++++
 1 files changed, 9 insertions(+), 0 deletions(-)

diff --git a/net/ipv4/tcp_cubic.c b/net/ipv4/tcp_cubic.c
index b758db1..a170778 100644
--- a/net/ipv4/tcp_cubic.c
+++ b/net/ipv4/tcp_cubic.c
@@ -270,6 +270,15 @@ static inline void bictcp_update(struct bictcp *ca, u32 cwnd)
 		ca->cnt = 100 * cwnd;              /* very small increment*/
 	}
 
+	/*
+	 * The initial growth of cubic function may be too conservative
+	 * when the available bandwidth is still unknown.
+	 */
+	if (ca->loss_cwnd == 0) {
+		if (ca->cnt > 20)	/* increase cwnd 5% per RTT */
+			ca->cnt = 20;
+	}
+
 	/* TCP Friendly */
 	if (tcp_friendliness) {
 		u32 scale = beta_scale;
-- 
1.7.0.4

Re: [PATCH 3/3] tcp_cubic: fix low utilization of CUBIC with HyStart

[Sangtae Ha]

Hi Lucas,

Thank you for reporting this problem and the results below show what
happened with low resolution clocks.
The detailed testing setup and results at
http://netsrv.csc.ncsu.edu/wiki/index.php/HyStart_testing_between_Princeton_and_NCSU
I will add the results in long rtt paths soon.

Sangtae

On Sun, Mar 13, 2011 at 7:04 PM, Sangtae Ha <sangtae.ha@gmail.com> wrote:
> HyStart sets the initial exit point of slow start.
> Suppose that HyStart exits at 0.5BDP in a BDP network and no history exists.
> If the BDP of a network is large, CUBIC's initial cwnd growth may be
> too conservative to utilize the link.
> CUBIC increases the cwnd 20% per RTT in this case.
>
> Signed-off-by: Sangtae Ha <sangtae.ha@gmail.com>
> ---
>  net/ipv4/tcp_cubic.c |    9 +++++++++
>  1 files changed, 9 insertions(+), 0 deletions(-)
>
> diff --git a/net/ipv4/tcp_cubic.c b/net/ipv4/tcp_cubic.c
> index b758db1..a170778 100644
> --- a/net/ipv4/tcp_cubic.c
> +++ b/net/ipv4/tcp_cubic.c
> @@ -270,6 +270,15 @@ static inline void bictcp_update(struct bictcp *ca, u32 cwnd)
>                ca->cnt = 100 * cwnd;              /* very small increment*/
>        }
>
> +       /*
> +        * The initial growth of cubic function may be too conservative
> +        * when the available bandwidth is still unknown.
> +        */
> +       if (ca->loss_cwnd == 0) {
> +               if (ca->cnt > 20)       /* increase cwnd 5% per RTT */
> +                       ca->cnt = 20;
> +       }
> +
>        /* TCP Friendly */
>        if (tcp_friendliness) {
>                u32 scale = beta_scale;
> --
> 1.7.0.4
>
>

Re: [PATCH 3/3] tcp_cubic: fix low utilization of CUBIC with HyStart

[Lucas Nussbaum]

On 13/03/11 at 19:07 -0500, Sangtae Ha wrote:
> Hi Lucas,
> 
> Thank you for reporting this problem and the results below show what
> happened with low resolution clocks.
> The detailed testing setup and results at
> http://netsrv.csc.ncsu.edu/wiki/index.php/HyStart_testing_between_Princeton_and_NCSU
> I will add the results in long rtt paths soon.

Thanks. What do you use to generate those graphs?

The graphs from HyStart(fixed) are a bit strange. Did HyStart really
detect something? It seems that it exited slow start due to losses, not
due to detection, since cwnd is decreased.
-- 
| Lucas Nussbaum             MCF Université Nancy 2 |
| lucas.nussbaum@loria.fr         LORIA / AlGorille |
| http://www.loria.fr/~lnussbau/  +33 3 54 95 86 19 |

Re: [PATCH 3/3] tcp_cubic: fix low utilization of CUBIC with HyStart

[Sangtae Ha]

On Mon, Mar 14, 2011 at 2:28 AM, Lucas Nussbaum <lucas.nussbaum@loria.fr> wrote:
> On 13/03/11 at 19:07 -0500, Sangtae Ha wrote:
>> Hi Lucas,
>>
>> Thank you for reporting this problem and the results below show what
>> happened with low resolution clocks.
>> The detailed testing setup and results at
>> http://netsrv.csc.ncsu.edu/wiki/index.php/HyStart_testing_between_Princeton_and_NCSU
>> I will add the results in long rtt paths soon.
>
> Thanks. What do you use to generate those graphs?

I used net/ipv4/tcp_probe.c and gnuplot to generate the graphs.

>
> The graphs from HyStart(fixed) are a bit strange. Did HyStart really
> detect something? It seems that it exited slow start due to losses, not
> due to detection, since cwnd is decreased.

You can see green points which indicate that ssthresh is set before the loss,
and right after that, the cwnd increases 20% per RTT since the
available bandwidth is still unknown.
This logic incurs far less number of losses compared to slow start and
hence more throughput.

> --
> | Lucas Nussbaum             MCF Université Nancy 2 |
> | lucas.nussbaum@loria.fr         LORIA / AlGorille |
> | http://www.loria.fr/~lnussbau/  +33 3 54 95 86 19 |
>

Sangtae