netdev
[Top] [All Lists]

Re: NAPI, e100, and system performance problem

To: Greg Banks <gnb@xxxxxxx>
Subject: Re: NAPI, e100, and system performance problem
From: Andi Kleen <ak@xxxxxx>
Date: Fri, 22 Apr 2005 13:36:56 +0200
Cc: Arthur Kepner <akepner@xxxxxxx>, "Brandeburg, Jesse" <jesse.brandeburg@xxxxxxxxx>, netdev@xxxxxxxxxxx, davem@xxxxxxxxxx
In-reply-to: <20050419055535.GA12211@xxxxxxx> (Greg Banks's message of "Tue, 19 Apr 2005 15:55:35 +1000")
References: <C925F8B43D79CC49ACD0601FB68FF50C03A633C7@orsmsx408> <Pine.LNX.4.61.0504180943290.15052@xxxxxxxxxx> <1113855967.7436.39.camel@xxxxxxxxxxxxxxxxxxxxx> <20050419055535.GA12211@xxxxxxx>
Sender: netdev-bounce@xxxxxxxxxxx
User-agent: Gnus/5.110002 (No Gnus v0.2) Emacs/21.3 (gnu/linux)
Greg Banks <gnb@xxxxxxx> writes:
>
> An inordinate amount of CPU is being spent running around polling the
> device instead of dealing with the packets in IP, TCP and NFS land.
> By inordinate, we mean twice as much or more cpu% than a MIPS/Irix
> box with slower CPUs.

We have seen similar behaviour. With NAPI some benchmarks run
a lot slower than on a driver on the same hardware/NIC without NAPI.
This can be even observed with simple tests like netperf single stream
between two boxes.

There seems to be also some problems with bidirectional traffic, although
I have not fully tracked them down to NAPI yet.

There is definitely some problem in NAPI land ...

-Andi

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