netdev
[Top] [All Lists]

Re: netdev.stats change suggestion

To: "David S. Miller" <davem@xxxxxxxxxx>
Subject: Re: netdev.stats change suggestion
From: Andi Kleen <ak@xxxxxxx>
Date: Thu, 24 Jan 2002 17:05:35 +0100
Cc: jgarzik@xxxxxxxxxxxxxxxx, ak@xxxxxxx, Robert.Olsson@xxxxxxxxxxx, cw@xxxxxxxx, dima@xxxxxxxxx, netdev@xxxxxxxxxxx
In-reply-to: <20020124.074729.41631242.davem@xxxxxxxxxx>
References: <20020124.062650.66057933.davem@xxxxxxxxxx> <20020124162825.A24611@xxxxxxxxxxxxx> <3C502A99.4EEFFB40@xxxxxxxxxxxxxxxx> <20020124.074729.41631242.davem@xxxxxxxxxx>
Sender: owner-netdev@xxxxxxxxxxx
User-agent: Mutt/1.3.22.1i
On Thu, Jan 24, 2002 at 07:47:29AM -0800, David S. Miller wrote:
>    From: Jeff Garzik <jgarzik@xxxxxxxxxxxxxxxx>
>    Date: Thu, 24 Jan 2002 10:39:05 -0500
>    
>    I have very little preference for the interface, besides -not-
>    dumping yet another random file into procfs...
> 
> Since netlink is available always now, let's use that.

Advantage of netlink is that it can block -- with /proc polling is always 
needed.
e.g. one could add setting of reporting thresholds to the interface and let 
netlink send a message when the counter overflows it. This way a statistic 
gathering tool could sleep and only wake up when something interesting happens.

[it's a real problem - when you have a whole gnome or windowmaker panel 
of statistics reporting tools around it can chew up a not insignificant amount 
of CPU time because they all wake up regularly and check /proc if nothing
has changed] 

Problem is only that sending rtnetlink messages from hard interrupt context
does not work very well, but it can be handled by using queue_event().

-Andi

<Prev in Thread] Current Thread [Next in Thread>