netdev
[Top] [All Lists]

Re: Tx queueing

To: Donald Becker <becker@xxxxxxxxx>
Subject: Re: Tx queueing
From: Andi Kleen <ak@xxxxxx>
Date: Thu, 18 May 2000 20:07:51 +0200
Cc: netdev@xxxxxxxxxxx
In-reply-to: <Pine.LNX.4.10.10005181222570.1658-100000@xxxxxxxxxxxxx>; from Donald Becker on Thu, May 18, 2000 at 06:36:32PM +0200
References: <392407D4.BE586507@xxxxxxxxxx> <Pine.LNX.4.10.10005181222570.1658-100000@xxxxxxxxxxxxx>
Sender: owner-netdev@xxxxxxxxxxx
On Thu, May 18, 2000 at 06:36:32PM +0200, Donald Becker wrote:
> > Has anyone done any serious work with NIC/CPU bonding?
> 
> The Mindcraft "benchmark" is superficially obvious, but the big network
> difference was that they were apparently using the TCP/IP checksum hardware
> on the i82559.  This has far more effect on SMP performance than anything
> else that was done.  We didn't even find out that the chip had the feature
> until months later, and still don't have the documentation on how to use
> it.

The Intel e100 driver implicitely documents it in C source. They unfortunately
use a very ugly method to implement it (own UDP/IP header parsing instead 
of CHECKSUM_HW). I was even porting it to your driver, until I discovered 
that my eepro100 had a far too old chip to implement it (i82557 stepping 1).

BTW, newer eepro100 driver versions like to often flood the network (collision 
light straight red) with bogus packets. The interrupts triggered
on the other system on the same segment lead to near solid livelock. 

That didn't happen with the older driver in straight 2.2.13.

-Andi
-- 
This is like TV. I don't like TV.

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