netdev
[Top] [All Lists]

Re: netdev.stats change suggestion

To: kuznet@xxxxxxxxxxxxx
Subject: Re: netdev.stats change suggestion
From: Jeff Garzik <garzik@xxxxxxxxxxxxx>
Date: Fri, 25 Jan 2002 13:33:08 -0500
Cc: Martin Devera <devik@xxxxxx>, davem@xxxxxxxxxx, netdev@xxxxxxxxxxx
In-reply-to: <200201251819.VAA27489@xxxxxxxxxxxxx>; from kuznet@xxxxxxxxxxxxx on Fri, Jan 25, 2002 at 09:19:14PM +0300
References: <Pine.LNX.4.10.10201242214140.15162-100000@xxxxxxxxxxxx> <200201251819.VAA27489@xxxxxxxxxxxxx>
Sender: owner-netdev@xxxxxxxxxxx
User-agent: Mutt/1.2.5i
On Fri, Jan 25, 2002 at 09:19:14PM +0300, kuznet@xxxxxxxxxxxxx wrote:
> > I can't found ifstat, nstat in my iproute 010824 .. Maybe old one ?
> 
> Yes, if appeared in the nest snapshot together with tcpdiag.
> 
> > By the way I don't see how you get over wraparound problem.
> 
> It needs sampling each sevral seconds to calculate rates,
> so that the problem just does not exist. 
> 
> 
> > in read stat part ? Fast, simple (and possibly wrong) :)
> 
> Yes. And this is exactly which I do not want to pay in addition
> to storing in the kernel information which it never wants to remember.

On the general topic of 64-bit counters, two facts weigh very heavily,
first, current and future hardware stores stats in 64-bit numbers, and
second, there are indeed MiBs which have 64-bit stats.  We had the
specific info at the kernel summit meeting... I will see if I can dig it
up.

If we do -not- have a way for NICs to dump 64-bit stats, each driver is
going to have scale stats down to 32 bits, and perhaps implement
in-driver sampling code.

Finally, IMHO the world is moving to machines that store 64-bit
numbers naturally and atomically.  Don't some of the new IA32 CPUs
even have support for 64-bit integers?

So, I beg pardon for being lost in the thread and not having found the
perfect spot to reply, but there it is :)

Regards,

        Jeff



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