-------Original Message-------
From: Tom Brinkman <[EMAIL PROTECTED]>
Sent: 08/11/03 04:33 AM
To: [EMAIL PROTECTED]
Subject: Re: [newbie] hdd copying VERY slow

> 
> On Sunday August 10 2003 06:21 pm, Joeb wrote:
> Just to add to this, if he's using /dev/hda and /dev/hdb aren't
> they both on the same IDE controller on most systems?  Since IDE
> can only write to one drive at a time per controller, improved
> performance would result in using the secondary IDE controller.
>  To further complicate it, since they're both on the same
> controller (and cable) and one is UDMA 133 and the other UDMA
> 100, the fastest they will go is 100 and maybe only 66 because of
> the mismatch (although I'm not positive on the 66 part).
>  Finally, I believe Windows is caching it's writes which will
> speed things up, at the expense of safety.
>
> Joeb

     There's two parts involved, the drive firmware (controller 
interface) and the controller on the motherboard. So unless you've 
got separate PCI controller cards, all the drives use the one 
controller, whether they're on the same cable or not. What is to be 
avoided is putting a CD drive and HDD on the same cable. In any 
event, no matter the number of controller cards you might have, 
they all use the one old and tired 33mhz PCI bus. There's a new PCI 
spec comin, PCI eXpress with a 66mhz bus. Like the AGP gimmick tho, 
it only marginally (+5%) improves performance.

>[root /tom] $ hdparm -tT /dev/hd[ab]
>
>/dev/hda:  (ata/133, udma6)
> Timing buffer-cache reads:   1232 MB in  2.00 seconds = 616.00 
>MB/sec
> Timing buffered disk reads:  140 MB in  3.00 seconds =  46.67 
>MB/sec
>
>/dev/hdb:  (ata/100, udma6)
> Timing buffer-cache reads:   1252 MB in  2.00 seconds = 626.00 
>MB/sec
> Timing buffered disk reads:  122 MB in  3.02 seconds =  40.40 
>MB/sec

      Both hda and hdb are on ide0. I have a cdrom and a burner on 
ide1. BTW I've got a week old motherboard. Aopen AK77-400 Max/n, 
KT400a chipset, IDE controller. It has 3 ide ports, ide0, ide1, 
ide2 and a Serial/ATA port. So I could'a separated the HDD drives, 
one to a cable (ide port). But I knew there's no point in that.  I 
believe the hdparm numbers back me up on that opinion ;) OTOH, 
hdparm spits out burst numbers. In the real world I get about 
20mb/sec transfers, ReiserFS to RieserFS, whether just moving files 
on the same HDD, or hda to hdb.

     I didn't wanna try separating the HDD's. When I installed the 
new mobo/cpu/ram, I configured the ide's just as they were on the 
old mobo. So when I booted up for the first time, my existing 
Mandrake 9.2 install found the new onboard NIC, and then went on 
like nothin was amiss.  If I had Windoze, I'd still be installing 
new drivers ;)  The latest cooker 2.4.22 kernel has support for the 
Serial/ATA port, but I don't have a S/ATA drive to try out.
-- 
    Tom Brinkman                  Corpus Christi, Texas


> 

Some correct me if I'm wrong (and I'm sure someone on this list will :) ).  But with 
two drives on a single IDE controller, hdparm will never show a degredation of speed 
because it tests each drive seperately, one after the other.  As such, at the time of 
testings, only one drive is being accessed so there is no contention.  If that is how 
the drives are used in real life, one at a time, then no problem, but for instance, 
let's say hda contains your web site and hdb has your ftp site and both have heavy 
access (of course, this is hypothetical, because if they are heavily accessed, you 
probably should be using SCSI).  Anyway, if that is the scenario, IDE0 can only talk 
to hda OR hdb, not both so you will introduce a contention.  If the hdb drive is moved 
to IDE1 (thus becoming hdc), then both hda and hdc can handle requests simultaneously.

On a more practical note, I do a lot of compiling of software.  To keep things simple, 
I have a source volume on a separate drive,  and there is a noticable difference when 
the source I'm compiling is on a different controller than the lib and headers of the 
actual linux distro.  Maybe it's just a coincidence, but it is significant in the 
amount of time difference (some things seconds, others minutes to hours).

Anyway, for normal use, it probably won't make much of a difference, but if you're 
trying to squeeze maximum throughput out of a system and you have the extra IDE 
channel available, why not use it?

Joeb


Want to buy your Pack or Services from MandrakeSoft? 
Go to http://www.mandrakestore.com

Reply via email to