netdev - RE: [PATCH] sch_red: fix weighted average calculation

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:	Fri, 14 Sep 2012 13:01:12 +0000
From:	"Dowdal, John" <jdowdal@...com>
To:	Eric Dumazet <eric.dumazet@...il.com>,
	"Chemparathy, Cyril" <cyril@...com>
CC:	"linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
	"netdev@...r.kernel.org" <netdev@...r.kernel.org>,
	"davem@...emloft.net" <davem@...emloft.net>,
	"david.ward@...mit.edu" <david.ward@...mit.edu>,
	"paul.gortmaker@...driver.com" <paul.gortmaker@...driver.com>
Subject: RE: [PATCH] sch_red: fix weighted average calculation

Eric, thank you for reviewing the code.  I now see the problem with the patch since backlog is an integer and qavg is a fixed point number at logW.  

We are considering another patch to update the comments to this code (with the actual C code change reverted) to stop the FPP by showing the derivation of the equation in the comments.  Does this sound good?


-----Original Message-----
From: Eric Dumazet [mailto:eric.dumazet@...il.com] 
Sent: Thursday, September 13, 2012 9:54 AM
To: Chemparathy, Cyril
Cc: linux-kernel@...r.kernel.org; netdev@...r.kernel.org; davem@...emloft.net; david.ward@...mit.edu; Dowdal, John; paul.gortmaker@...driver.com
Subject: Re: [PATCH] sch_red: fix weighted average calculation

On Thu, 2012-09-13 at 09:43 -0400, Cyril Chemparathy wrote:
> This patch fixes an apparent bug in the running weighted average calculation
> used in the RED algorithm.
> 
> Going by the described formula:
> 	   qavg = qavg*(1-W) + backlog*W
> 	=> qavg = qavg + (backlog - qavg) * W
> 
> ... with W converted to a pre-calculated shift, this then becomes:
> 	qavg = qavg + (backlog - qavg) >> logW
> 
> ... giving the modified expression introduced by this patch.
> 
> Signed-off-by: John Dowdal <jdowdal@...com>
> ---
>  include/net/red.h |    2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/include/net/red.h b/include/net/red.h
> index ef46058..05960a4 100644
> --- a/include/net/red.h
> +++ b/include/net/red.h
> @@ -287,7 +287,7 @@ static inline unsigned long red_calc_qavg_no_idle_time(const struct red_parms *p
>  	 *
>  	 * --ANK (980924)
>  	 */
> -	return v->qavg + (backlog - (v->qavg >> p->Wlog));
> +	return v->qavg + (backlog - v->qavg) >> p->Wlog;
>  }
>  
>  static inline unsigned long red_calc_qavg(const struct red_parms *p,

This is going to be a FPP (Frequently Posted Patch)

Current formulae is fine.

Thats because backlog, at start of red_calc_qavg_no_idle_time() is not
yet scaled by p->Wlog. v->avg is scaled, but not backlog.

Have you tested RED after your patch ?