HD speeds - normal?

pickarooney

Hi,

Continuing the ongoing investigation into the drunk-badger-in-glue-like performance of my machine, I ran the hdparm command on my hard-drives, with what seems to be fairly unimpressive results.

I have a new (3 months old) 80 GB Western Digital hard drive, 7200rpm 8MB cache as master (hdb), and a slightly older (1.5 years) Maxtor 40GB, 5600rpm as slave(hda).
The results of hdparm -tT look pretty dire in comparison with those I've seen posted elsewhere.

For the WD 80GB:

/dev/hdb1:
Timing buffer-cache reads: 128 MB in 1.17 seconds =109.40 MB/sec
Timing buffered disk reads: 64 MB in 3.35 seconds = 19.10 MB/sec

/dev/hdb6:
Timing buffer-cache reads: 128 MB in 1.17 seconds =109.40 MB/sec
Timing buffered disk reads: 64 MB in 3.40 seconds = 18.82 MB/sec

/dev/hdb:
multcount = 16 (on)
IO_support = 1 (32-bit)
unmaskirq = 1 (on)
using_dma = 1 (on)
keepsettings = 0 (off)
readonly = 0 (off)
readahead = 8 (on)
geometry = 9729/255/63, sectors = 156301488, start = 0



For the Maxtor 40GB

/dev/hda1:
Timing buffer-cache reads: 128 MB in 1.19 seconds =107.56 MB/sec
Timing buffered disk reads: 64 MB in 3.23 seconds = 19.81 MB/sec

/dev/hda:
multcount = 16 (on)
IO_support = 1 (32-bit)
unmaskirq = 1 (on)
using_dma = 1 (on)
keepsettings = 0 (off)
readonly = 0 (off)
readahead = 8 (on)
geometry = 4982/255/63, sectors = 80043264, start = 0

My CD-burner on the secondary IDE controller only burns at 8x instead of 24x as advertised.

I tried swapping all the IDE devices around in several configurations, with no discernible improvement.

Have I IDE problems or are these figures reasonable for a fairly modest system?

I know this is not strictly a unix issue, but it might make more sense to people here given the context.

Cheers,
P

flamegrill

P4 2.4GHz
1GB DDR ram
2x80GB 7200rpm 8mb cache disks:

hdparm -tT /dev/md0

/dev/md0:
Timing buffer-cache reads: 128 MB in 0.46 seconds =278.26 MB/sec
Timing buffered disk reads: 64 MB in 1.23 seconds = 52.03 MB/sec


Obviously using software Raid would have an impact. But none the less its not a bad performance.

Dell Poweredge 2240
666mhz PIII
256mb Ram
3x9.1GB ultra 160 SCSI disks

/usr/sbin/hdparm -tT /dev/sda

/dev/sda:
Timing buffer-cache reads: 128 MB in 0.61 seconds =209.84 MB/sec
Timing buffered disk reads: 64 MB in 1.62 seconds = 39.51 MB/sec

Those are modest results.

Have you got DMA enabled?

Paul

pickarooney

Of course I forgot to post my specs *slap*

800MHz Duron, 320MB RAM.

AFAIK dma is enabled, I get a return of using_dma = 1 (on) from the hdparm command without arguments and with i as switch I get

PIO modes: pio0 pio1 pio2 pio3 pio4
DMA modes: mdma0 mdma1 mdma2
UDMA modes: udma0 udma1 *udma2 udma3 udma4 udma5

Having run a couple of tests, 110 is the highest I can get for buffer-cache reads, straight after reboot, and it immediately drops to 80ish if I start up a p2p program (normal enough, with all the HD hammering). Compared with your results, mine are not as wojus as compared with the 550ish bcr values I saw (the chap was complaining about slow HD access in the thread). Would it be worth my while getting 80-thingy IDE cables? Should I save up for a new computer and save everyone the hassles of reading my near-daily outpourings of technocrank angst? Should Batman reveal his barely-concealed lust for the Boy Wonder and be done with it?

P

flamegrill

Heh - your going to hate me:

XP 2600
1GB 333MHZ DDR
2x120GB 8MB 7200rpm drives


/dev/hdb:
Timing buffer-cache reads: 1304 MB in 2.00 seconds = 652.00 MB/sec
Timing buffered disk reads: 140 MB in 3.01 seconds = 46.51 MB/sec

/dev/hda:
Timing buffer-cache reads: 1288 MB in 2.00 seconds = 644.00 MB/sec
Timing buffered disk reads: 144 MB in 3.04 seconds = 47.37 MB/sec

Both disks are on the same controller, and on the same cable. DMA is enabled.

Quite impressive me thinks.

Paul

pickarooney

"Hate" is such a small word..

I'm going to try some new IDE cables for the fun of it, but I reckon the beast has had its day, and droid-rot has well and truly set in.

longword

Originally posted by flamegrill
2x120GB 8MB 7200rpm drives
Both disks are on the same controller, and on the same cable. DMA is enabled.

Why would you do such a thing? It's madness! Since the kernel doesn't implement disconnected operation or tagged command queues on ATA (nobody does yet AFAIK), only one drive can ever be active at any given time on the wire so you're effectively limiting your transfer rate to little more than one drive. Plug them into different IDE channels, even if one has to share with a slave CDROM drive.

And I thought I was bad plugging my five 10kRPM SCSI disks into the one U160 cable...

pickarooney

Originally posted by longword
Why would you do such a thing? It's madness! Since the kernel doesn't implement disconnected operation or tagged command queues on ATA (nobody does yet AFAIK), only one drive can ever be active at any given time on the wire so you're effectively limiting your transfer rate to little more than one drive. Plug them into different IDE channels, even if one has to share with a slave CDROM drive.

And I thought I was bad plugging my five 10kRPM SCSI disks into the one U160 cable...

I have no idea what the bit beginning with "tagged command queues..." means, but I was always taught to put fast IDE devices together and slow ones together to maximise data transfer between hard disks, at least in Windows. Are the rules all different for linux? If so that would be messy for dual-booting types.

Gerry

ibm ( now hitachi ) drives support tagged command queuing. I saw an option in the 2.6 kernel config to enable it, but that machine didn't have an ibm drive..
must do some testing on this machine in freebsd..

Typedef

I'd say this is something to do with your mobo possibly, assuming you don't have data operations going on on hdx, when doing the test and hdy and hdy, isn't broken somehow, it has to be mobo or possibly a pos[1] ide cable.

Also, maybe you should check your kernel config, for obvious foobars.

[1] Piece of shit.

nadir

2.4 Ghz P4
FSB 533 Mhz
256MB 333MHZ DDR


/dev/hdb:
Timing buffer-cache reads: 1528 MB in 2.00 seconds = 764.00 MB/sec
Timing buffered disk reads: 126 MB in 3.00 seconds = 41.93 MB/sec

multcount = 16 (on)
IO_support = 0 (default 16-bit)
unmaskirq = 0 (off)
using_dma = 1 (on)
keepsettings = 0 (off)
readonly = 0 (off)
readahead = 8 (on)
geometry = 119150/16/63, sectors = 120103200, start = 0

I need more memory

pickarooney

One obvious foobar is the row of 99 99 99 that appears during LILO booting. Is there a quick way to fix that?
Also, how does one update a motherboard chipset in linux?

niallb

Hi P,
your machine is pretty similar spec to the one I use all day.
(I have something faster, but it sounds like a helicopter,
and is only powered up for deadlines.)
Mine's a PIII 650 (Fastest I could find that didn't need a fan)
and similar RAM (392M)
In normal usage, I'm not swapping at all,
so unless you're running a load of services,
your disk should be available for better access than you're seeing.

I don't use my IDE controller at all though I remember getting
about 15M/s on my newest drive.
I have a 60G drive (halfway between your vintages) on the IDE bus on
the faster machine - which is also an AMD based system.)
I'm not using the optimal cable, just a good quality old ribbon,
and this is what it gives. Plenty fast enough for my needs.

/dev/hda:
Timing buffer-cache reads: 128 MB in 0.66 seconds =193.94 MB/sec
Timing buffered disk reads: 64 MB in 2.22 seconds = 28.83 MB/sec

It's a good 50% faster than your results,
and the chipset it's using is
VIA Technologies|VT82C586 IDE (Apollo)
I have
CONFIG_BLK_DEV_VIA82CXXX=y
CONFIG_IDEDMA_AUTO=y
set in the kernel config.

If your mobo runs a Duron, how well matched is your kernel?
I don't notice you naming your IDE controller.
Are you using a kernel compiled for Duron or just a standard OS install?
Post the output of lspci and check which IDE controller you have.
When my Athlon boots, I get:
VP_IDE: VIA vt8235 (rev 00) IDE UDMA133 controller on pci00:11.1
in my logs.

Unless you get similar, you're probably only using standard IDE support.
If you have via sound and/or via ethernet,
there's a good chance that you can benefit from this.

Compiling in the right driver here doubled my throughput IIRC.

This is worth pursuing,
If I've enough details, I'd fire together a kernel for you if you like,
but it's worth building it yourself as you're bound to
see more tweaks you can make along the way!

It should at least get rif of the glue, and even a
drunk badger can get into some surprising places...

NiallB

pickarooney

Hi Niall,

The output of lspci is as follows, got more via than an Aran Island postal address:

00:00.0 Host bridge: VIA Technologies, Inc. VT8363/8365 [KT133/KM133] (rev 03)
00:01.0 PCI bridge: VIA Technologies, Inc. VT8363/8365 [KT133/KM133 AGP]
00:07.0 ISA bridge: VIA Technologies, Inc. VT82C686 [Apollo Super South] (rev 40)
00:07.1 IDE interface: VIA Technologies, Inc. VT82C586/B/686A/B PIPC Bus Master IDE (rev 06)
00:07.2 USB Controller: VIA Technologies, Inc. USB (rev 16)
00:07.3 USB Controller: VIA Technologies, Inc. USB (rev 16)
00:07.4 Bridge: VIA Technologies, Inc. VT82C686 [Apollo Super ACPI] (rev 40)
00:11.0 Ethernet controller: Realtek Semiconductor Co., Ltd. RTL-8139/8139C/8139C+ (rev 10)
00:13.0 Multimedia audio controller: C-Media Electronics Inc CM8738 (rev 10)
01:00.0 VGA compatible controller: ATI Technolgies Inc Radeon RV100 QY [Radeon 7000/VE]

Bus Master IDE looks like a match for your config.
I am unable to answer any questions relating to the kernal, as I didn't make it myself, it's Mandrake 9.1's bog-standard 2.4.21-0.13mdk.
A quick grep for VP_IDE in the /var/logs directory produces:

dmesg:VP_IDE: IDE controller at PCI slot 00:07.1
dmesg:VP_IDE: chipset revision 6
dmesg:VP_IDE: not 100% native mode: will probe irqs later
dmesg:VP_IDE: VIA vt82c686b (rev 40) IDE UDMA100 controller on pci00:07.1

My sound card is an integraed CM and my network card a realtek (PCI).

Is that in any way informative?

Thanks a million
P

niallb

Very informative,
could you also put up the outputs of
lsmod
cat /proc/interrupts and
cat /proc/dma
and let me know what your root partition is (hda1,hda5 or whatever).
You can attach if it gets too longwinded!

Take a look for the via file in proc.
cat /proc/ide/via
If its NOT there, you're not running VIA support.

If nothing is there, TRY running
modprobe via82cxxx.

I'VE NEVER RUN THIS AS A MODULE,
ONLY COMPILED IN AT BOOT TIME.
BE CAREFUL!
IF YOU SMELL BURNING TURN IT OFF :-)

Try it, you might get better results.
I still think you're better off recompiling the kernel,
but if you add the line
via82cxxx
to /etc/modules Mandrake will stick it in for you on boot.

If you get an 80 pin cable, you can add the line
ide0=ata66 to /etc/lilo.conf to enable that.
(Don't be fooled by the speed '66',
I think all it's saying is "Look Ma! I've got a blue cable")

I'll check back later, and see if your machine is smokin...

NiallB

pickarooney

/proc/ide/via is a 0 byte empty file.

Cause I'm not all that smart or brave, should I literally type modprobe via82cxxx or should the Xs be replaced by more meaningful symbols such as VT82C686?

/etc/modules contains one uncommented line:
scsi_hostadapter
Does it matter where the other line is added (before, after this)?
I still don't feel comfortable with recompiling kernels, btw

Here's the output of those 3 logs, in order.

Cheers,
P

niallb

via82cxxx is the name of the driver.
It detects the specific chipset when it loads,
No need to change its name.

Did you try lsmod?
The via may be already loaded, but not working.
I have no /proc/ide/via file on the machine I'm not using
IDE drives with.

Run a locate via82cxxx.o beforehand,
quite possibly the Mandrake 9.1 kernel didn't build it.
If it is there, modprobe should stick it in.

/etc/modules in mandrake is used to load modules at boot time.
The entry you see for scsi_hostadapter will have a corresponding alias in /etc/modules.conf

NiallB

pickarooney

Strange, my last post disappeared...

Anyway, that module wasn't anywhere on my system, so I donwloaded it from here

I copied it into /lib/modules/2.4.21-0.13mdk/kernel/drivers/ide then ran a locate -u. Modprobe couldn't find it. I then copied it into a directory in the root account's path, tried modprobe again, still couldn't find it. Where does it need to be copied?

BTW, the output of lsmod is at the beginning of the text file in my last post.

Cheers,
P

niallb

That module almost certainly won't work,
and even if it seems to, I'd be very wary.

If you want to use it, you should use the
whole kernel from the floppyfw distribution.
You can't easily mix and match kernel and modules.

You'll need a via82cxxx.o from the 2.4.2.21-0.13mdk
kernel or a new and matching kernel.

I'll stick a link up to one later in the day.

NiallB

pickarooney

That module almost certainly won't work

[Mrs. Doyle]Oooh! Well! (hysterical laughter) I certainly won't be doing that then!
[/Mrs. Doyle]

Is there such a thing as an idiot's guide to (re)compiling a kernel? I've never read a guide from start to finish that didn't lose me about half way through.

If I could get the hang of it I could maybe improve my gfx card performance at the same time.

Cheers,
P

niallb

You'll win all round on performance,
and gain some IDE stability improvements as well.

If you have broadband, do yourself a favour
and get the latest kernel as one file.
2.4.23 is just out.

wget http://ftp.esat.net:/pub/linux/kernel/v2.4/linux-2.4.23.tar.bz2

The patches won't apply cleanly to your 2.4.21*mdk tree,
so it's easier to use the vanilla one.
Any third party patches and modules would expect this anyway.

cd /usr/src
ls
mv linux linux-2.4.21-0.13mdk if necessary
You might as well keep the mandrake tree in case things
go pear shaped.

tar jxvf linux-2.4.23.tar.bz2 -C /usr/src
cd /usr/src/linux

Download http://niallb.lapsedhippy.com/config-Athlon-Duron-23
and copy it to /usr/src/linux/.config
This will build in the via ide support, so you won't need a module.

One change I would make if I were you is this:
cd /usr/src/linux
cp Makefile Makefile-2.4.23 you'll need to replace this later to patch to 2.4.24 cleanly if required
vim Makefile
find the line starting EXTRAVERSION =
and change it to
EXTRAVERSION = -pickarooney1
This will make things much simpler with multiple versions,
and puts your name in the kernel.

Now run make xconfig to see what your settings are,

I've set sensible values as far as I can tell for your system,
so you'll get a good idea of what you need and what you don't.
I built in support for your ethernet and soundcard,
so you'll need to change things if you want to use alsa.
This is mainly in case you don't get modules right first time.
I also set you up for VIA AGP and Radeon support, which should improve X.
Get used to building the kernel a few times before you
go compiling alsa for it.
I'm running alsa-1.0rc3 at the moment which is great,
but the final 1.0 should be out very soon.


Compiling without changes should give you a usable system.
When you exit xconfig (Personally I prefer make menuconfig try it too) you need to run:
make dep ; make clean ; make bzImage && make modules && make modules_install
Looks worse than it is :-)
Now if you look in /lib/modules you should see a new directory
called /lib/modules/2.4.23-pickarooney1

Your kernelfile should now be in /usr/src/linux/arch/i386/boot/bzImage

Copy the kernelfile to /boot/vmlinuz-2.4.23-pickarooney1
and make an entry in /etc/lilo.conf or /boot/grub/grub.conf
Usually you want the system map and the .config there too.
cp /usr/src/linux/arch/i386/boot/bzImage /boot/vmlinuz-2.4.23-pickarooney1
cp /usr/src/System.map /boot/System.map-2.4.23-pickarooney1
cp /usr/src/linux/.config /boot/config-2.4.23-pickarooney1

edit /etc/lilo.conf
and copy a stanza.
Edit it to read:
image=/boot/vmlinuz-2.4.23-pickarooney1
label="Duron"
root=/dev/hda5 or whatever
append= "whatever append was working for you."
If you got the high speed cables, add ide0=ata66 to the append line.

That's it in a nutshell,
tell us how you get on,
NiallB

P.S. I don't have access to your actual hardware,
so it might not work or it might DO BAD THINGS,
but it seems pretty close to me!

pickarooney

Hi Niall,

I didn't get a chance to do it last night, I should be able this evening. Do I need to backup anything apart from the /usr/src/linux directory?

P

niallb

Hi P,
you don't really even need to back that up,
as you have it in an RPM.
Changing your kernel shouldn't cause you any damage,
but I suppose it can, so at least get /home offside.

It's quite possible your first kernel won't succesfully boot,
but just select your previous known good one from the
boot menu and try again.
If it all goes wrong, Mandrake 9.1 CD1 has a rescue mode that will reinstall the boot loader.

Take your time though, and back up all your data if you have the space..
Next, make sure you can restore it :-),
then you can do other things.

Use whatever backup utility you like.
I just use something liketar czvf /BACKUPSPACE/home.tar.gz /home
and dump to tape or CD.
For a regular backup, rsync is great, but use the --numeric-ids option between systems.

As a rule of thumb, you should be backing up
/home
/usr/local Keep anything not in RPMs here if you can
/etc Well, the configuration work is all here.
/var/spool In case you've got yet to be delivered mail or print jobs.
/var/log so you see what used to happen before it broke
/root if you've got essential ssh keys or config files
/boot & /lib/modules Your working kernel and friends.

Good luck,
NiallB

pickarooney

Failed at step one. I don't have a /usr/src/linux directory. The only thing in /usr/source is the RPM dir.
Does this matter?

Continuning on, I got as far as
make dep ; make clean ; make bzImage && make modules && make modules_install
but that seemed to go wrong. The last few output lines are:

make[2]: Leaving directory `/usr/src/linux-2.4.23/kernel'
make[1]: *** [_sfdep_kernel] Error 2
make[1]: Leaving directory `/usr/src/linux-2.4.23'
make: *** [dep-files] Error 2

Nothing new was created in /lib/modules

I tried piping the command through tee -a to create a text file log, but it didn't create.
I stopped there for now as I didn't see the point in going any further with errors.

niallb

Hi,
You've no /usr/src/linux directory simply
because you didn't install the kernel source before.

Nothing new was created in /lib/modules because the make modules never ran.
the && in the line make dep ; make clean ; make bzImage && make modules && make modules_install
means only run the next command if the previous command was succesful. The ; simply seperate commands.
(BTW, you can use || to run a command only if the previous one 'fails' - exitcode != 0 )

I wish I could see some of the lines before the ones below,
to see what part of the kernel is failing.
Possibly, it's something minor that you could leave out of the config. The .config builds fine here though.
More likely, it might be a piece of kit you don't have installed that's needed to compile the kernel.
Try using urpmi --test kernel-source to install the 2.4.21 kernel source, and see what other packages it says you need.
Then, install those and try again.
You'll need at least:
glibc-devel
ncurses-devel


If you want to make a logfile, you could seperate the
above command sequence onto seperate lines.
Your 'tee' command was probably waiting for the output
of 'make modules_install' which never happened.
For example make bzImage | tee logfile
will get you more mileage.
More usually you can use the script command.
(part of util-linux).
Just run script /tmp/kernelcompile_20031210T2320.log or whatever springs to mind.
Anything you do will then be logged to that file.
Type ^D or exit to finish logging.

Good luck on your next attempt,

NiallB

pickarooney

Try 2:

I checked for glibc-devel and ncurses-devel - both present and correct.
I ran urpmi --test kernel-source which downloaded the .rpm and tested it correctly with no missing dependencies. (side note, it downloaded it from a US site as I never added my installation CDs back to rpm's source list, but I reckon it was quicker than going upstairs and finding the CDs )

I ran the make commands again and logged them in the attached script log. It really didn't get very far into the process, but it's all gobbledigook to me.

niallb

Looks like you're missing some bits alright,
you have make, gcc, glibc-devel, ncurses-devel.
How about binutils?

I ran off a copy of that kernel if you want to take a look,
though building it yourself will allow you tweak it.

http://niallb.lapsedhippy.com/pickarooney-kernel_2.4.23-niallb1.tgz

Download and extract it to your root directory with tar zxvf

We've got to find out if it makes things faster :-)

NiallB

pickarooney

Hi niall, just a quick reply before I head out to work:

locate binutils
/usr/share/doc/binutils-2.13.90.0.18
/usr/share/doc/binutils-2.13.90.0.18/README
/usr/share/doc/libbinutils2-2.13.90.0.18
/usr/share/doc/libbinutils2-2.13.90.0.18/README
/usr/share/info/binutils.info-1.bz2
/usr/share/info/binutils.info-2.bz2
/usr/share/info/binutils.info-3.bz2
/usr/share/info/binutils.info.bz2

pickarooney

OK, i unpacked that kernel in the root directory.

niallb

Any Luck?
Have you had a chance to add it to lilo yet?

pickarooney

Not much luck, I lost sound, mouse, some hard drives on booting with the new kernel, and the hdparm results are identical to before.
I'm still trying to figure out what the compilation problem is with the modified kernel...

niallb

You lost sound because you're probably using alsa.
I couldn't easily build alsa as well for you, so I built
in the OSS drivers instead.
It would require some small changes to use that instead.
The config was meant to just be a starting point anyway :-)

How you lost mouse, I don't know.
What kind of a mouse do you use?

How you lost some hard drives as well doesn't make
immediate sense to me.
Do you have other controllers besides the one you mentioned?
(I only built in support for the one you named).

Did you check dmesg for mentions of the VIA driver?
What does hdparm say?

NiallB

flamegrill

Lads take it to IRC or something surely ?

Paul

niallb

If you say so.

If you don't have a VIA motherboard, you don't have to read it,
and if you do, you might be interested.

It had though gotten a bit long winded :-)

NiallB

pickarooney

Just a final note before the thread is closed.

The hard drives didn't actually get lost, they were undetected then redetected.
The pointing-stick is a cordless optical usb mouse, easy enough to reconfigure (if you have another mouse handy )
hdparm -tT gave results of 107-110

In the heel of the hunt, as a strange uncle of mine used to say, there's no improvement, but alsa no hdparm done, and I reckon I learned something (no idea what, but at one stage I got the distinct feeling I knew what I was doing).

Thanks a million for your help and patience Niall
P