What is your speed of disks in Gentoo

DZello · Posted: Tue Apr 19, 2005 7:01 pm Post subject:

Raising the PCI clock speed when overclocking can decrease your hard disk performance. I've got this problem with an onboard Promise IDE controller.
_________________
PC (Gentoo): Athlon 64 @2.5 Ghz, A8V, 1 GB ram, ATI 9600XT, 160 + 250 GB HDD
Laptop (ArchLinux): IBM Thinkpad R40, Intel Centrino 1.4 GHz, 512 MB ram, 40 GB HDD
Kernel: 2.6.19-Beyond2
Linux apostle since 1998

Rumil · Posted: Wed Apr 20, 2005 6:39 am Post subject:

Lucky B · Posted: Wed Apr 20, 2005 7:21 am Post subject:

Here's my system so you can compare

Ard Righ · Posted: Wed Apr 20, 2005 8:49 am Post subject:

Does hdparm support SATA drives yet ? I cannot enable hdparm, because it doesn't like my two WD Raptor SATA drives
_________________
Answer a simple question with a simple answer!

zieloo · Veteran Joined: 28 Mar 2004 Posts: 1337

Gentree · Posted: Wed Apr 20, 2005 10:59 am Post subject:

Rumil · Posted: Wed Apr 20, 2005 11:30 am Post subject:

Gentree · Posted: Wed Apr 20, 2005 12:08 pm Post subject:

ext3 ->R4

zieloo · Veteran Joined: 28 Mar 2004 Posts: 1337

So he should be happy with his 7 or 8mb/sec...

Gentree · Posted: Wed Apr 20, 2005 12:41 pm Post subject:

Gentree · Posted: Wed Apr 20, 2005 12:47 pm Post subject:

@ Lucky B

Gentree · Posted: Wed Apr 20, 2005 12:59 pm Post subject:

Lucky B · Posted: Wed Apr 20, 2005 3:18 pm Post subject:

Gentree · Posted: Wed Apr 20, 2005 4:22 pm Post subject:

Thanks,

Is there any reason you mount NTFS rw? Last time I looked the kernel ntfs support could only write if the file was exactly the same size!

In any case if you use linux ntfs it does not mean much (ie anything at all) as a comparison of read or write performance of NTFS.

What HD do you have , its odd that you have about the same for buffer reads as my Barracuda but nearly 4x faster on cached reads.

_________________
Linux, because I'd rather own a free OS than steal one that's not worth paying for.
Gentoo because I'm a masochist
AthlonXP-M on A7N8X. Portage ~x86

Lucky B · Posted: Wed Apr 20, 2005 5:48 pm Post subject:

Gentree · Posted: Wed Apr 20, 2005 7:00 pm Post subject:

hdtach ? I'd like something better than hdparm to assess hd's.

[EDIT] Well I found hdtach, a windows only freeware , but it seems not to do much more than hdparm from what I see. Only does sequencial reads from the disk.

I have not tried it since I dont have a working win installation atm.

Why do you say it is "deeper"?
Thx

_________________
Linux, because I'd rather own a free OS than steal one that's not worth paying for.
Gentoo because I'm a masochist
AthlonXP-M on A7N8X. Portage ~x86

Lucky B · Posted: Fri Apr 22, 2005 8:10 pm Post subject:

I say it's deeper because it takes into account random reads through the the hard drive (measuring speed as you fall away from the spindle) not just random accesses.

Also, about your tests, I don't think they're 100% accurate or very fair to compare to mine. Copying many files is not always sequential in nature so you're basically copying sparse files, this puts undue emphasis on the filesystem hierarchy more than the drive interface itself. (I know pagefile.sys is one big chunk because I created it that way

)

I will do further tests tonight by repartitioning my FAT32 partitions to include an NTFS, ext3, reiser, r4, xfs and fat32 partition in each drive so that I can test all 36 combinations and report on it.

EDIT: actually I think there are 42 combinations total but I am not gonna do some of em =P
_________________
We are Dyslexic of Borg. Fusistance is retile. Your ass will be laminated.

Gentree · Posted: Sun Apr 24, 2005 9:01 am Post subject:

OK, if you are sure hdtach does random reads I stand corrected , like I said , I did not actually install it because I dont have any win* at the moment. I had read that it only did sequencial reads but that may no have been reliable.

As for my test not being as "fair" as yours , I think we both indicated clearly what we were doing so both are as valid but tested different situations.

Your test on a big file will make it sequencial and will get a "best case" result for the fs under test. Which is fine and valid as such but something I hardy ever will do in reality.

What is more important to me is a slightly more "realistic use" test which is why I chose to time something I actually did .

The capability of a fs to create and delete a large number if variously sized (often small) files is an important factor in assessing what fs I am going to use.

Some fs will perform very well on big files but equally poorly on a "many small file" copy/move and vice versa.

What would interest me is to see some timings from a native NTFS copy under windoze and see how it compares to a similar copy under Linux. This would have to be stopwatch timings rather that the completely unrealistic data speeds shown by windows during copying.

BTW I think to test all poss combination of 6 filing systems you will have 6x5x4x3x2 tests to do . Good luck.

I dont think fat32 is very interesting since it is an non-journalling fs so you could poss divide you workload by a factor of 6 by dropping it.

The only real use I see for fat32 is as a common rw partition for a dual boot system, in which case it is the only choise so no need to compare timings.

Anyway good luck and I look forward to seeing your results.

_________________
Linux, because I'd rather own a free OS than steal one that's not worth paying for.
Gentoo because I'm a masochist
AthlonXP-M on A7N8X. Portage ~x86

Lucky B · Posted: Sun Apr 24, 2005 5:14 pm Post subject:

actually it's 6x6x2 6x6 matrix for one drive then compare using two drives. And I can probably script it using nested loops.

I haven't done the test yet because I accidentally neutered my partitions when adding a new logical partition. Quick fix but didn't feel like testing anything afterwards =P
_________________
We are Dyslexic of Borg. Fusistance is retile. Your ass will be laminated.

superboy2k · n00b Joined: 30 Jan 2004 Posts: 17

I've the same problem!! I tried everything change all hardware (mb, nic)
The pc is a 440Bx board, 256Mb, Celeron 450, ultrastard 10k disk:

/dev/sda:
Timing cached reads: 480 MB in 2.01 seconds = 238.25 MB/sec
Timing buffered disk reads: 104 MB in 3.03 seconds = 34.35 MB/sec

this is ok, but if i make a file transfert or a network one, throughput is ~5 Mb/s, why?
In win i get 10Mb/s of network transfer and more in filesystem one.

it seams like pci performances sucks, i don't know what is it! i've tried 2.6.11 gentoo and mm...

kornhs4 · Posted: Wed Apr 27, 2005 11:11 am Post subject: Asus 3800N

I also had a disk-speed on hda approx. 5Mb/s. After starting udma mode and rc-update add hdparm boot I get:

/dev/hda:
Timing cached reads: 2280 MB in 2.00 seconds = 1138.47 MB/sec
Timing buffered disk reads: 82 MB in 3.04 seconds = 26.98 MB/sec

/dev/hda:

Model=IC25N060ATMR04-0, FwRev=MO3OAD4A, SerialNo=MRG31YKCJPKB4H
Config={ HardSect NotMFM HdSw>15uSec Fixed DTR>10Mbs }
RawCHS=16383/16/63, TrkSize=0, SectSize=0, ECCbytes=4
BuffType=DualPortCache, BuffSize=7884kB, MaxMultSect=16, MultSect=16
CurCHS=16383/16/63, CurSects=16514064, LBA=yes, LBAsects=117210240
IORDY=on/off, tPIO={min:240,w/IORDY:120}, tDMA={min:120,rec:120}
PIO modes: pio0 pio1 pio2 pio3 pio4
DMA modes: mdma0 mdma1 mdma2
UDMA modes: udma0 udma1 *udma2
AdvancedPM=yes: mode=0x80 (128) WriteCache=enabled
Drive conforms to: ATA/ATAPI-6 T13 1410D revision 3a:

/etc/conf.d/hdparm:

# Copyright 1999-2004 Gentoo Foundation
# Distributed under the terms of the GNU General Public License v2
# $Header: /var/cvsroot/gentoo-x86/sys-apps/hdparm/files/hdparm-conf.d.3,v 1.2 2004/09/06 02:17:08 swegener Exp $

# You can either set hdparm arguments for each drive using hdX_args,
# discX_args, cdromX_args and genericX_args, e.g.
#
hda_args="-d1 -c1 -u1 -m16 -X66"
# disc1_args="-d1"
cdrom0_args="-d1 -c1 -u1 -X66"

# or, you can set hdparm options for ALL drives using all_args, e.g.
#
#all_args="-d1 -c1 -u1 -m16 -x66"

My System is a Asus Laptop A3800N, Centrino 1,6Ghz...
_________________
_________________
Life would be easier
if i had the source code

opentaka · l33t Joined: 18 Feb 2005 Posts: 840 Location: Japan

superboy2k · n00b Joined: 30 Jan 2004 Posts: 17

Zyne · Guru Joined: 08 Jun 2004 Posts: 334

#pfo · n00b Joined: 25 Apr 2005 Posts: 16 Location: vienna