ICH5 with 2.6.1 very slow

ICH5 with 2.6.1 very slow

[Kyle]

Problem with ICH5?
I read the list web, so please CC my email, thanks.

P4 2.6G HT, 2GB Ram, ICH5, WD250GB/8M x 2  (md raid 1), kernel 2.6.1
Timing buffer-cache reads: 128 MB in 0.14 seconds =882.89 MB/sec
Timing buffered disk reads: 64 MB in 2.14 seconds = 29.93 MB/sec

Celeron 1.3T, 1GB Ram, SIS630/5513, WD80GB/2M x 2 (md raid 1), kernel 2.6.1
Timing buffer-cache reads: 128 MB in 1.24 seconds =103.08 MB/sec
Timing buffered disk reads: 64 MB in 1.83 seconds = 35.00 MB/sec

Celeron 2.5G, 512MB Ram, i845/ICH4, WD120GB/8M x 2 (md raid 1), redhat
kernel 2.4.20
Timing buffer-cache reads:   128 MB in  0.35 seconds =365.71 MB/sec
Timing buffered disk reads:  64 MB in  1.38 seconds = 46.38 MB/sec

Kyle

Re: ICH5 with 2.6.1 very slow

[Bas Mevissen]

Kyle wrote:

> Problem with ICH5?
> I read the list web, so please CC my email, thanks.
> 
> P4 2.6G HT, 2GB Ram, ICH5, WD250GB/8M x 2  (md raid 1), kernel 2.6.1
> Timing buffer-cache reads: 128 MB in 0.14 seconds =882.89 MB/sec
> Timing buffered disk reads: 64 MB in 2.14 seconds = 29.93 MB/sec
> 
> Celeron 1.3T, 1GB Ram, SIS630/5513, WD80GB/2M x 2 (md raid 1), kernel 2.6.1
> Timing buffer-cache reads: 128 MB in 1.24 seconds =103.08 MB/sec
> Timing buffered disk reads: 64 MB in 1.83 seconds = 35.00 MB/sec
> 
> Celeron 2.5G, 512MB Ram, i845/ICH4, WD120GB/8M x 2 (md raid 1), redhat
> kernel 2.4.20
> Timing buffer-cache reads:   128 MB in  0.35 seconds =365.71 MB/sec
> Timing buffered disk reads:  64 MB in  1.38 seconds = 46.38 MB/sec
> 

Is that a parallel IDE? I own a S-ATA IDE (also WD) with an ICH5. That
performs around 45-50MB/sec with 2.6.1. I've never tried 2.4.xx on it.

Bas.

Re: ICH5 with 2.6.1 very slow

[Kyle]

Yes they are PATA, I expect something like 40-50MB/s, now my much slower
Celeron 1.3T with 80GB/2M perform better than my ICH5!

Kyle

----- Original Message ----- 
From: "Bas Mevissen" <ml@basmevissen.nl>
To: "Kyle" <kyle@southa.com>
Cc: <linux-kernel@vger.kernel.org>
Sent: Friday, February 06, 2004 1:23 AM
Subject: Re: ICH5 with 2.6.1 very slow


> Kyle wrote:
>
> > Problem with ICH5?
> > I read the list web, so please CC my email, thanks.
> >
> > P4 2.6G HT, 2GB Ram, ICH5, WD250GB/8M x 2  (md raid 1), kernel 2.6.1
> > Timing buffer-cache reads: 128 MB in 0.14 seconds =882.89 MB/sec
> > Timing buffered disk reads: 64 MB in 2.14 seconds = 29.93 MB/sec
> >
> > Celeron 1.3T, 1GB Ram, SIS630/5513, WD80GB/2M x 2 (md raid 1), kernel
2.6.1
> > Timing buffer-cache reads: 128 MB in 1.24 seconds =103.08 MB/sec
> > Timing buffered disk reads: 64 MB in 1.83 seconds = 35.00 MB/sec
> >
> > Celeron 2.5G, 512MB Ram, i845/ICH4, WD120GB/8M x 2 (md raid 1), redhat
> > kernel 2.4.20
> > Timing buffer-cache reads:   128 MB in  0.35 seconds =365.71 MB/sec
> > Timing buffered disk reads:  64 MB in  1.38 seconds = 46.38 MB/sec
> >
>
> Is that a parallel IDE? I own a S-ATA IDE (also WD) with an ICH5. That
> performs around 45-50MB/sec with 2.6.1. I've never tried 2.4.xx on it.
>
> Bas.
>
>
>

Re: ICH5 with 2.6.1 very slow

[Bas Mevissen]

Kyle wrote:

> Yes they are PATA, I expect something like 40-50MB/s, now my much slower
> Celeron 1.3T with 80GB/2M perform better than my ICH5!
> 

What does 'hdparm /dev/hdX' say (X=a,b,c...)?

Bas.

Re: ICH5 with 2.6.1 very slow

[Kyle]

hdparm /dev/hda
/dev/hda:
multcount = 16 (on)
IO_support = 1 (32-bit)
unmaskirq = 1 (on)
using_dma = 1 (on)
keepsettings = 0 (off)
readonly = 0 (off)
readahead = 256 (on)
geometry = 30401/255/63, sectors = 488397168, start = 0

tried with hdparm -a8192 /dev/hda, not much different
/dev/hdc same as /dev/hda

Kyle
----- Original Message ----- 
From: "Bas Mevissen" <ml@basmevissen.nl>
To: "Kyle" <kyle@southa.com>
Cc: <linux-kernel@vger.kernel.org>
Sent: Friday, February 06, 2004 1:30 AM
Subject: Re: ICH5 with 2.6.1 very slow


> Kyle wrote:
>
> > Yes they are PATA, I expect something like 40-50MB/s, now my much slower
> > Celeron 1.3T with 80GB/2M perform better than my ICH5!
> >
>
> What does 'hdparm /dev/hdX' say (X=a,b,c...)?
>
> Bas.
>
>
>

Re: ICH5 with 2.6.1 very slow

[Bas Mevissen]

Kyle wrote:

> hdparm /dev/hda
> /dev/hda:
> multcount = 16 (on)
> IO_support = 1 (32-bit)
> unmaskirq = 1 (on)
> using_dma = 1 (on)
> keepsettings = 0 (off)
> readonly = 0 (off)
> readahead = 256 (on)
> geometry = 30401/255/63, sectors = 488397168, start = 0
> 
> tried with hdparm -a8192 /dev/hda, not much different
> /dev/hdc same as /dev/hda
> 

Looks fine. That's strange. Maybe reboot the system and check 
/proc/interrupts to see if none of them is excessive.

What's your .config? Try with as less as possible IDE stuff enabled.

Bas.

Re: ICH5 with 2.6.1 very slow

[Kyle]

Today I tried to boot with redhat kernel 2.4.20-24.9smp, much better result:

/dev/hda:
 Timing buffer-cache reads:   128 MB in  0.14 seconds =914.29 MB/sec
 Timing buffered disk reads:  64 MB in  1.67 seconds = 38.32 MB/sec

Really strange that /dev/hdc is even faster!
/dev/hdc:
 Timing buffer-cache reads:   128 MB in  0.13 seconds =984.62 MB/sec
 Timing buffered disk reads:  64 MB in  1.17 seconds = 54.70 MB/sec

I ran hdparm 5 times at both harddisk and take the fastest result.
/dev/hda and /dev/hdc are both WD 250GB / 8M, same model.

at kernel 2.6, I got same performance for  /dev/hda and /dev/hdc, around
~28MB/sec. 50%-100% slower then kernel 2.4

/proc/interrupt:

           CPU0       CPU1
  0:    8014632          0    IO-APIC-edge  timer
  1:       2273          0    IO-APIC-edge  i8042
  2:          0          0          XT-PIC  cascade
  8:          1          0    IO-APIC-edge  rtc
  9:          0          0   IO-APIC-level  acpi
 14:     167888          0    IO-APIC-edge  ide0
 15:     199399          0    IO-APIC-edge  ide1
 20:    2186070          0   IO-APIC-level  eth1
NMI:          0          0
LOC:    8014888    8014887
ERR:          0
MIS:          0

I tried to update my mobo BIOS (Intel D865GLC), remove one of the NIC (I
have two), disable serial, parallel, USB etc just leave IDE and built-in
display, no help. I have NO IDE cdrom and floppy, only 2 harddisk, 1 CPU, 4
x 512M DDR and a mobo at my machine.

CONFIG_IDE=y
CONFIG_BLK_DEV_IDE=y
CONFIG_BLK_DEV_IDEDISK=y
CONFIG_IDEDISK_MULTI_MODE=y
CONFIG_BLK_DEV_IDECD=m
CONFIG_BLK_DEV_IDESCSI=m
CONFIG_BLK_DEV_IDEPNP=y
CONFIG_BLK_DEV_IDEPCI=y
CONFIG_IDEPCI_SHARE_IRQ=y
CONFIG_BLK_DEV_GENERIC=y
CONFIG_BLK_DEV_IDEDMA_PCI=y
CONFIG_IDEDMA_PCI_AUTO=y
CONFIG_BLK_DEV_ADMA=y
CONFIG_BLK_DEV_PIIX=y
CONFIG_BLK_DEV_IDEDMA=y
CONFIG_IDEDMA_AUTO=y

Kyle



----- Original Message ----- 
From: "Bas Mevissen" <ml@basmevissen.nl>
To: "Kyle" <kyle@southa.com>
Cc: <linux-kernel@vger.kernel.org>
Sent: Friday, February 06, 2004 6:10 PM
Subject: Re: ICH5 with 2.6.1 very slow


> Kyle wrote:
>
> > hdparm /dev/hda
> > /dev/hda:
> > multcount = 16 (on)
> > IO_support = 1 (32-bit)
> > unmaskirq = 1 (on)
> > using_dma = 1 (on)
> > keepsettings = 0 (off)
> > readonly = 0 (off)
> > readahead = 256 (on)
> > geometry = 30401/255/63, sectors = 488397168, start = 0
> >
> > tried with hdparm -a8192 /dev/hda, not much different
> > /dev/hdc same as /dev/hda
> >
>
> Looks fine. That's strange. Maybe reboot the system and check
> /proc/interrupts to see if none of them is excessive.
>
> What's your .config? Try with as less as possible IDE stuff enabled.
>
> Bas.
>
>
>

Re: ICH5 with 2.6.1 very slow

[Bas Mevissen]

Kyle wrote:

> Today I tried 

(...)

This is quite strange. The only thing I can think of is that the 
hardware (?) raid1 is causing problems with 2.6.

Is there a possibility for you to test without it?

Currently, I don't have a decent PATA disk luyng around, so I cannot 
verify anything for you.

Regards,

Bas.

Re: ICH5 with 2.6.1 very slow

[Kyle Wong]

No hardware raid, it's kernel built-in md driver.

----- Original Message ----- 
From: "Bas Mevissen" <ml@basmevissen.nl>
To: "Kyle" <kyle@southa.com>
Cc: <linux-kernel@vger.kernel.org>
Sent: Monday, February 09, 2004 4:32 PM
Subject: Re: ICH5 with 2.6.1 very slow


> Kyle wrote:
> 
> > Today I tried 
> 
> (...)
> 
> This is quite strange. The only thing I can think of is that the 
> hardware (?) raid1 is causing problems with 2.6.
> 
> Is there a possibility for you to test without it?
> 
> Currently, I don't have a decent PATA disk luyng around, so I cannot 
> verify anything for you.
> 
> Regards,
> 
> Bas.
> 
> 
> 

Re: ICH5 with 2.6.1 very slow

[Kyle]

Today I tried with compile the kernel 2.6.1 with:

IGNORE word93 Validation BITS (IDEDMA_IVB) = y

The result looks a bit better, got 30MB/s at /dev/hda and 37MB/s at /dev/hdc
(38MB/s and 55MB/s at kernel 2.4.20)

Still very strange, /dev/hda and /dev/hda are exactly same model harddisk.

Also, the result still can't compare with my another much slower machine
Celeron / ICH4 / 2 x WD 120GB (md1), which got 46MB/s at both /dev/hda and
/dev/hdc,

Kyle

----- Original Message ----- 
From: "Bas Mevissen" <ml@basmevissen.nl>
To: "Kyle" <kyle@southa.com>
Cc: <linux-kernel@vger.kernel.org>
Sent: Monday, February 09, 2004 4:32 PM
Subject: Re: ICH5 with 2.6.1 very slow


> Kyle wrote:
>
> > Today I tried
>
> (...)
>
> This is quite strange. The only thing I can think of is that the
> hardware (?) raid1 is causing problems with 2.6.
>
> Is there a possibility for you to test without it?
>
> Currently, I don't have a decent PATA disk luyng around, so I cannot
> verify anything for you.
>
> Regards,
>
> Bas.
>
>
>

Re: ICH5 with 2.6.1 very slow

[Jamie Lokier]

Kyle wrote:
> Today I tried with compile the kernel 2.6.1 with:
> 
> IGNORE word93 Validation BITS (IDEDMA_IVB) = y
> 
> The result looks a bit better, got 30MB/s at /dev/hda and 37MB/s at /dev/hdc
> (38MB/s and 55MB/s at kernel 2.4.20)

Aha...

Have a look at the thread called "[RFC] IDE 80-core cable detect -
chipset-specific code to over-ride eighty_ninty_three()".

It specifically deals with ICH5 and is probably the same problem as
you're seeing.

-- Jamie

Re: ICH5 with 2.6.1 very slow

[Kyle]

Yes I read the thread and re-compile with IDEDMA_IVB= y, got some
performance back, but is still far slower than 2.4

----- Original Message ----- 
From: "Jamie Lokier" <jamie@shareable.org>
To: "Kyle" <kyle@southa.com>
Cc: "Bas Mevissen" <ml@basmevissen.nl>; <linux-kernel@vger.kernel.org>
Sent: Thursday, February 12, 2004 4:41 PM
Subject: Re: ICH5 with 2.6.1 very slow


> Kyle wrote:
> > Today I tried with compile the kernel 2.6.1 with:
> >
> > IGNORE word93 Validation BITS (IDEDMA_IVB) = y
> >
> > The result looks a bit better, got 30MB/s at /dev/hda and 37MB/s at
/dev/hdc
> > (38MB/s and 55MB/s at kernel 2.4.20)
>
> Aha...
>
> Have a look at the thread called "[RFC] IDE 80-core cable detect -
> chipset-specific code to over-ride eighty_ninty_three()".
>
> It specifically deals with ICH5 and is probably the same problem as
> you're seeing.
>
> -- Jamie
>

Re: ICH5 with 2.6.1 very slow

[Bas Mevissen]

Jamie Lokier wrote:
>
> Have a look at the thread called "[RFC] IDE 80-core cable detect -
> chipset-specific code to over-ride eighty_ninty_three()".
> 
> It specifically deals with ICH5 and is probably the same problem as
> you're seeing.
> 

Ive already pointed Kyle to that thread. But the udma settings seem to 
be OK.

So my guess now is that this is not just an issue with the driver only.

Regards,

Bas.

Re: ICH5 with 2.6.1 very slow

[Jeff Garzik]

Another idea...  check and make sure your hardware is set to "enhanced 
mode" in BIOS.  "legacy mode" and "combined mode" would definitely be slow.

	Jeff

Re: ICH5 with 2.6.1 very slow

[Kyle]

It's already in "Enhanced mode", tried "legacy" and "combined" as well, no
help.

----- Original Message ----- 
From: "Jeff Garzik" <jgarzik@pobox.com>
To: "Bas Mevissen" <ml@basmevissen.nl>
Cc: "Jamie Lokier" <jamie@shareable.org>; "Kyle" <kyle@southa.com>;
<linux-kernel@vger.kernel.org>
Sent: Friday, February 13, 2004 3:28 AM
Subject: Re: ICH5 with 2.6.1 very slow


> Another idea...  check and make sure your hardware is set to "enhanced
> mode" in BIOS.  "legacy mode" and "combined mode" would definitely be
slow.
>
> Jeff
>
>
>
>
>

Re: ICH5 with 2.6.1 very slow

[Daniel Blueman]

Did you try replacing the two IDE cables you have with new 30cm 80 way
ones? IDE can hide marginal cables with CRC-checking and retrying.

What about swapping them, and/or swapping the drives around, then
comparing the transfer rate of /dev/hda and /dev/hdc again?

"Kyle" <kyle@southa.com> wrote in message
news:<1oClr-6j-15@gated-at.bofh.it>...
> It's already in "Enhanced mode", tried "legacy" and "combined" as well, no
> help.

-- 
Daniel J Blueman

GMX ProMail (250 MB Mailbox, 50 FreeSMS, Virenschutz, 2,99 EUR/Monat...)
jetzt 3 Monate GRATIS + 3x DER SPIEGEL +++ http://www.gmx.net/derspiegel +++

Re: ICH5 with 2.6.1 very slow

[Kyle]

today since one of my mirrored harddisk (/dev/hda) failed, I removed it from
md-raid1 and now /dev/hdc becomes /dev/hda

hdparm -t /dev/hda gets me ~37MB/s now (before: /dev/hda - 30MB/s,
/dev/hdc - 37MB/s)

maybe there's problem with /dev/hda so it's relatively slower!

However, the result still much slower than kernel 2.4.20 (55MB/s)

Kyle

Re: ICH5 with 2.6.1 very slow

[Bartlomiej Zolnierkiewicz]

On Sunday 15 of February 2004 19:48, Kyle wrote:
> today since one of my mirrored harddisk (/dev/hda) failed, I removed it
> from md-raid1 and now /dev/hdc becomes /dev/hda
>
> hdparm -t /dev/hda gets me ~37MB/s now (before: /dev/hda - 30MB/s,
> /dev/hdc - 37MB/s)
>
> maybe there's problem with /dev/hda so it's relatively slower!
>
> However, the result still much slower than kernel 2.4.20 (55MB/s)

Please fill bugzilla entry (htp://bugzilla.kernel.org)
and attach 'dmesg' and 'lspci -vvv -xxx' outputs for 2.4.20 and 2.6.x.

It would be also helpful to narrow down the issue to kernel version when
this slowdown started (2.4.20 -> 2.6.x means too much changes to anybody
sane to even start thinking about going through all of them).

--bart

Re: ICH5 with 2.6.1 very slow

[Martin Schlemmer]

[-- Attachment #1: Type: text/plain, Size: 991 bytes --]

On Sun, 2004-02-15 at 21:19, Bartlomiej Zolnierkiewicz wrote:
> On Sunday 15 of February 2004 19:48, Kyle wrote:
> > today since one of my mirrored harddisk (/dev/hda) failed, I removed it
> > from md-raid1 and now /dev/hdc becomes /dev/hda
> >
> > hdparm -t /dev/hda gets me ~37MB/s now (before: /dev/hda - 30MB/s,
> > /dev/hdc - 37MB/s)
> >
> > maybe there's problem with /dev/hda so it's relatively slower!
> >
> > However, the result still much slower than kernel 2.4.20 (55MB/s)
> 
> Please fill bugzilla entry (htp://bugzilla.kernel.org)
> and attach 'dmesg' and 'lspci -vvv -xxx' outputs for 2.4.20 and 2.6.x.
> 
> It would be also helpful to narrow down the issue to kernel version when
> this slowdown started (2.4.20 -> 2.6.x means too much changes to anybody
> sane to even start thinking about going through all of them).
> 

Also a hdparm -i /dev/hda might help I guess (as the small default
read-ahead causes this for many users)

-- 
Martin Schlemmer

[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 189 bytes --]

Re: ICH5 with 2.6.1 very slow

[Kyle]

Today I tried "lspci -vvv -xxx" with 2.6.1, forgotten to do this with
2.4.20, and I found this:

00:1f.1 IDE interface: Intel Corp. 82801EB ICH5 IDE (rev 02) (prog-if 8a
[Master SecP PriP])
        Subsystem: Intel Corp.: Unknown device 4c43
        Control: I/O+ Mem+ BusMaster+ SpecCycle- MemWINV- VGASnoop- ParErr-
Stepping- SERR- FastB2B-
        Status: Cap- 66Mhz- UDF- FastB2B+ ParErr- DEVSEL=medium >TAbort-
<TAbort- <MAbort- >SERR- <PERR-
        Latency: 0
        Interrupt: pin A routed to IRQ 18
        Region 0: I/O ports at <unassigned>
        Region 1: I/O ports at <unassigned>
        Region 2: I/O ports at <unassigned>
        Region 3: I/O ports at <unassigned>
        Region 4: I/O ports at ffa0 [size=16]
        Region 5: Memory at 7f800000 (32-bit, non-prefetchable) [size=1K]

00:1f.2 IDE interface: Intel Corp.: Unknown device 24d1 (rev 02) (prog-if 8f
[Master SecP SecO PriP PriO])
        Subsystem: Intel Corp.: Unknown device 4c43
        Control: I/O+ Mem- BusMaster+ SpecCycle- MemWINV- VGASnoop- ParErr-
Stepping- SERR- FastB2B-
        Status: Cap- 66Mhz+ UDF- FastB2B+ ParErr- DEVSEL=medium >TAbort-
<TAbort- <MAbort- >SERR- <PERR-
        Latency: 0
        Interrupt: pin A routed to IRQ 18
        Region 0: I/O ports at e800 [size=8]
        Region 1: I/O ports at e400 [size=4]
        Region 2: I/O ports at e000 [size=8]
        Region 3: I/O ports at dc00 [size=4]
        Region 4: I/O ports at d800 [size=16]

00:1d.2 USB Controller: Intel Corp. 82801EB USB (Hub #3) (rev 02) (prog-if
00 [UHCI])
        Subsystem: Intel Corp.: Unknown device 4c43
        Control: I/O+ Mem- BusMaster+ SpecCycle- MemWINV- VGASnoop- ParErr-
Stepping- SERR- FastB2B-
        Status: Cap- 66Mhz- UDF- FastB2B+ ParErr- DEVSEL=medium >TAbort-
<TAbort- <MAbort- >SERR- <PERR-
        Latency: 0
        Interrupt: pin C routed to IRQ 18
        Region 4: I/O ports at d000 [size=32]

[root@s1 kyle]# hdparm -i /dev/hda

/dev/hda:

 Model=WDC WD2500JB-00EVA0, FwRev=15.05R15, SerialNo=WD-WMAEH1328328
 Config={ HardSect NotMFM HdSw>15uSec SpinMotCtl Fixed DTR>5Mbs FmtGapReq }
 RawCHS=16383/16/63, TrkSize=57600, SectSize=600, ECCbytes=74
 BuffType=DualPortCache, BuffSize=8192kB, MaxMultSect=16, MultSect=16
 CurCHS=16383/16/63, CurSects=16514064, LBA=yes, LBAsects=268435455
 IORDY=on/off, tPIO={min:120,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 udma3 udma4 *udma5
 AdvancedPM=no WriteCache=enabled
 Drive conforms to: device does not report version:  1 2 3 4 5 6

Does it means that ICH5 shared IRQ 18 with USB and cause the performance
problem?
I'll try to submit this to bugzilla.

Kyle