xfs
[Top] [All Lists]

Re: XFS "freezes" on AMD systems

To: Florin Andrei <florin@xxxxxxx>
Subject: Re: XFS "freezes" on AMD systems
From: Vincent Janelle <random@xxxxxxxxxxxxxxxxx>
Date: Thu, 26 Sep 2002 17:01:46 -0400
Cc: linux-xfs@xxxxxxxxxxx
In-reply-to: <1033072528.28775.54.camel@xxxxxxxxxxxxxxxxxxx>
Organization: http://www.goblinstudios.com/
References: <15761.27018.659158.237109@xxxxxxxxxxxxxxxxxx> <20020925075844.GA26750@xxxxxxxxxxxxx> <15762.6020.364002.585622@xxxxxxxxxxxxxxxxxx> <3D9239C1.2070704@xxxxxxxxxxx> <20020925185258.664fece3.random@xxxxxxxxxxxxxxxxx> <1033072528.28775.54.camel@xxxxxxxxxxxxxxxxxxx>
Sender: linux-xfs-bounce@xxxxxxxxxxx
thats mostly due to command queing.. Unless you're using relatively new 
ultra-160 drives, and a good controller its still possible fill up the 
command queue relatively quickly with modern systems..  There are also 
'modern' IDE controllers out there like 3ware cards that fix quite a 
few issues like this (although they're primarily raid cards).

Only reason why I use scsi over IDE in my *servers* is that the warranty 
and MTBF are usually 3 years or more, whereas on IDE they're 1 year.

On 26 Sep 2002 13:35:28 -0700
Florin Andrei <florin@xxxxxxx> wrote:

> He's right, but you're right too. :-)
> A properly tuned IDE works fine usually.
> However, all other things being equal, a disk connected via SCSI "feels"
> faster than the same HDD gear connected via IDE. The reason being the
> way SCSI talks with the rest of the system (lower CPU usage, etc).
> 
> On Wed, 2002-09-25 at 15:52, Vincent Janelle wrote:
> > irrelevant and flamebait.
> > 
> > Proper tuning of DMA settings will help, as well as making sure that using 
> > the correct IDE driver is selected.
> > 
> > hdparm -c1 -d1 -u1 /dev/hd[a,b,c,etc] will turn on 32bit write support, 
> > dma, and unmasked irq's.  
> > 
> > man hdparm of course before doing any of this, YMMV.
> > 


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