netdev
[Top] [All Lists]

Re: [PATCH] use mmiowb in tg3_poll

To: Lennert Buytenhek <buytenh@xxxxxxxxxxxxxx>
Subject: Re: [PATCH] use mmiowb in tg3_poll
From: Arthur Kepner <akepner@xxxxxxx>
Date: Mon, 30 May 2005 09:30:13 -0700 (PDT)
Cc: netdev@xxxxxxxxxxx, jesse.barnes@xxxxxxxxx, gnb@xxxxxxx
In-reply-to: <20050528231209.GA8456@xxxxxxxxxxxxxxxxx>
References: <200410211628.06906.jbarnes@xxxxxxxxxxxx> <Pine.LNX.4.33.0410221345400.392-100000@xxxxxxxxxxxxxxxxxxxxx> <20050528231209.GA8456@xxxxxxxxxxxxxxxxx>
Sender: netdev-bounce@xxxxxxxxxxx
On Sun, 29 May 2005, Lennert Buytenhek wrote:

> .....
> I'm quite curious what kind of MMIO read latency you see on your
> Altix boxen.  This app is quite useful for determining those figures
> on x86{,_64} machines:
> 
>       http://svn.gnumonks.org/trunk/mmio_test/mmio_test.c
> ....

[cc list change: jbarnes@xxxxxxxxxxxx -> jesse.barnes@xxxxxxxxx]

Haven't tried this program yet (though it looks interesting.) 
In the past I've instrumented the driver with get_cycles() to 
acquire the PIO read and mmiowb() latencies.

I quickly looked through the records that I have and, unfortunately, 
wasn't able to locate raw data. But I found some scrawlings in my 
notes which say the most recent measurements for an Altix are: 

        i)  PIO read latency ~ 2.4 usec
        ii) mmiowb() latency ~ 1.1 usec

(This data was taken when the PIOs/mmiowb()s were coming from a CPU 
which was "nearest" to the NIC.)

If you have data for x86{,_64}, I'd be interested in that, too.

--
Arthur

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