[TCP] Fixed mss in tcp_init_cwnd

[TCP] Fixed mss in tcp_init_cwnd

[Herbert Xu]

[-- Attachment #1: Type: text/plain, Size: 360 bytes --]

Hi Dave:

The MSS in tcp_init_cwnd should be the real one instead of the TSO value.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

Cheers,
-- 
Visit Openswan at http://www.openswan.org/
Email: Herbert Xu ~{PmV>HI~} <herbert@gondor.apana.org.au>
Home Page: http://gondor.apana.org.au/~herbert/
PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt

[-- Attachment #2: p --]
[-- Type: text/plain, Size: 492 bytes --]

===== net/ipv4/tcp_input.c 1.73 vs edited =====
--- 1.73/net/ipv4/tcp_input.c	2004-09-13 10:30:58 +10:00
+++ edited/net/ipv4/tcp_input.c	2004-09-27 17:00:32 +10:00
@@ -799,10 +799,10 @@
 	__u32 cwnd = (dst ? dst_metric(dst, RTAX_INITCWND) : 0);
 
 	if (!cwnd) {
-		if (tp->mss_cache > 1460)
+		if (tp->mss_cache_std > 1460)
 			cwnd = 2;
 		else
-			cwnd = (tp->mss_cache > 1095) ? 3 : 4;
+			cwnd = (tp->mss_cache_std > 1095) ? 3 : 4;
 	}
 	return min_t(__u32, cwnd, tp->snd_cwnd_clamp);
 }

Re: [TCP] Fixed mss in tcp_init_cwnd

[Nivedita Singhvi]

Herbert Xu wrote:
> Hi Dave:
> 
> The MSS in tcp_init_cwnd should be the real one instead of the TSO value.
> 
> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
> 
> Cheers,
> 
> 
> ------------------------------------------------------------------------
> 
> ===== net/ipv4/tcp_input.c 1.73 vs edited =====
> --- 1.73/net/ipv4/tcp_input.c	2004-09-13 10:30:58 +10:00
> +++ edited/net/ipv4/tcp_input.c	2004-09-27 17:00:32 +10:00
> @@ -799,10 +799,10 @@
>  	__u32 cwnd = (dst ? dst_metric(dst, RTAX_INITCWND) : 0);
>  
>  	if (!cwnd) {
> -		if (tp->mss_cache > 1460)
> +		if (tp->mss_cache_std > 1460)
>  			cwnd = 2;
>  		else
> -			cwnd = (tp->mss_cache > 1095) ? 3 : 4;
> +			cwnd = (tp->mss_cache_std > 1095) ? 3 : 4;
>  	}
>  	return min_t(__u32, cwnd, tp->snd_cwnd_clamp);
>  }

Re: [TCP] Fixed mss in tcp_init_cwnd

[Nivedita Singhvi]

Herbert Xu wrote:

> ===== net/ipv4/tcp_input.c 1.73 vs edited =====
> --- 1.73/net/ipv4/tcp_input.c	2004-09-13 10:30:58 +10:00
> +++ edited/net/ipv4/tcp_input.c	2004-09-27 17:00:32 +10:00
> @@ -799,10 +799,10 @@
>  	__u32 cwnd = (dst ? dst_metric(dst, RTAX_INITCWND) : 0);
>  
>  	if (!cwnd) {
> -		if (tp->mss_cache > 1460)
> +		if (tp->mss_cache_std > 1460)
>  			cwnd = 2;
>  		else
> -			cwnd = (tp->mss_cache > 1095) ? 3 : 4;
> +			cwnd = (tp->mss_cache_std > 1095) ? 3 : 4;
>  	}
>  	return min_t(__u32, cwnd, tp->snd_cwnd_clamp);
>  }

Helps to send after finishing commenting..

I fixed this locally but was still seeing poor throughput,
though it does correct how many we initially/on restart send out.
Thought this was fixed in the bk tree?

thanks,
Nivedita

Re: [TCP] Fixed mss in tcp_init_cwnd

[David S. Miller]

On Mon, 27 Sep 2004 18:08:28 +1000
Herbert Xu <herbert@gondor.apana.org.au> wrote:

> The MSS in tcp_init_cwnd should be the real one instead of the TSO value.

This happens too early in the connections lifetime
to make a difference for anything.

It happens on the first move to ESTABLISHED state, and
therefore by definition before we try to send any data
on the connection.

You're barking up the wrong tree in the search for these
TSO problems.  Instead, I'd recommend looking seriously at
the tcp_clean_rtx_queue() partial ack'ing stuff I spoke
about yesterday.

Re: [TCP] Fixed mss in tcp_init_cwnd

[Herbert Xu]

On Mon, Sep 27, 2004 at 12:00:31PM -0700, David S. Miller wrote:
> On Mon, 27 Sep 2004 18:08:28 +1000
> Herbert Xu <herbert@gondor.apana.org.au> wrote:
> 
> > The MSS in tcp_init_cwnd should be the real one instead of the TSO value.
> 
> This happens too early in the connections lifetime
> to make a difference for anything.

I know.  That's why I put it in a different thread.

IMHO we should still fix it though at least for consistency.
-- 
Visit Openswan at http://www.openswan.org/
Email: Herbert Xu ~{PmV>HI~} <herbert@gondor.apana.org.au>
Home Page: http://gondor.apana.org.au/~herbert/
PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt

Re: [TCP] Fixed mss in tcp_init_cwnd

[David S. Miller]

On Tue, 28 Sep 2004 08:02:40 +1000
Herbert Xu <herbert@gondor.apana.org.au> wrote:

> On Mon, Sep 27, 2004 at 12:00:31PM -0700, David S. Miller wrote:
> > On Mon, 27 Sep 2004 18:08:28 +1000
> > Herbert Xu <herbert@gondor.apana.org.au> wrote:
> > 
> > > The MSS in tcp_init_cwnd should be the real one instead of the TSO value.
> > 
> > This happens too early in the connections lifetime
> > to make a difference for anything.
> 
> I know.  That's why I put it in a different thread.
> 
> IMHO we should still fix it though at least for consistency.

That early on in the connection, it should be setting both
values, right?  If so your patch still needs a tweak. :)

Re: [TCP] Fixed mss in tcp_init_cwnd

[Herbert Xu]

On Mon, Sep 27, 2004 at 04:04:54PM -0700, David S. Miller wrote:
>
> > IMHO we should still fix it though at least for consistency.
> 
> That early on in the connection, it should be setting both
> values, right?  If so your patch still needs a tweak. :)

I'm not saying that this patch makes any difference in terms of
run-time results.  At that point both mss_cache and mss_cache_std
should contain the same values.

However, since the value that's intended here is the physical MSS,
we should use mss_cache_std for the sake of consistency.

The only reason I spotted this at all is because I grepped -w for
mss_cache and found this function as one of the very few users.

Cheers,
-- 
Visit Openswan at http://www.openswan.org/
Email: Herbert Xu ~{PmV>HI~} <herbert@gondor.apana.org.au>
Home Page: http://gondor.apana.org.au/~herbert/
PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt