Tyan 2466 SMP locks hard with 2.4.20 + heavy disk i/o yet runs 2.2.* without problems

Tyan 2466 SMP locks hard with 2.4.20 + heavy disk i/o yet runs 2.2.* without problems

[Dirk Eddelbuettel]

System:
  Tyan 2466 MPX board, 2 Athlon 1800MP, bios 1.01 from March 2002
  1 gb ram, ibm + maxtor ide disks, ide cdrom + cdrw 
  ati rage 128, sb pci 128, on-board 3c905 nic
  no other cards
  hardware configuration unaltered since June of last year

Performance under 2.2.20 with ext3 + win4lin patches
  + Flawless
  + Uptime of _over 200 days_ with fairly heavy use until UPS got tripped
  + Can run bonnie++ with no issues other than very noticeable i/o slowdown

Performance under 2.4.20 with ext3 + win4lin patch
  - "Stable" until disk-heavy operation causes freeze, typically within 
    two to three days
  - Heavy disk use (full backup writing, diff against big tarball, bonnie++) 
    freeze the machine hard, no ping, no sign of live
  + memtest86 shows no problem with the ram

I searched via google and google groups, but found no pertinent idea for my
case:  No raid or scsi controller, memory doesn't seem to be an issue as it
works really, really well under 2.2.20 and memtest86 passes too. The system 
is generally very stable under 2.2.20 --- but gets killed by 2.4.20.

I run fairly stock kernels (from Debian kernel source packages) with the 
win4lin patches, and the ext3 patch in the case of 2.2.20.  ACPI is disabled.
Disk access is not tuned in any way, DMA gets enabled by default.

I would be happy to try out suggestions in terms of different kernel
configurations, or patches.  I would prefer to stay with the 2.4.* kernel for
now.

Thanks in advance for any pointers. CCs are welcome.

Dirk

-- 
Don't drink and derive. Alcohol and algebra don't mix.

Re: Tyan 2466 SMP locks hard with 2.4.20 + heavy disk i/o yet runs 2.2.* without problems

[Alan Cox]

On Iau, 2003-05-01 at 14:52, Dirk Eddelbuettel wrote:
> Performance under 2.4.20 with ext3 + win4lin patch
>   - "Stable" until disk-heavy operation causes freeze, typically within 
>     two to three days
>   - Heavy disk use (full backup writing, diff against big tarball, bonnie++) 
>     freeze the machine hard, no ping, no sign of live
>   + memtest86 shows no problem with the ram

If you don't have a PS/2 mouse plugged into the box add one. If that
doens't help duplicate the crash without win4lin.

Re: Tyan 2466 SMP locks hard with 2.4.20 + heavy disk i/o yet runs 2.2.* without problems

[Dirk Eddelbuettel]

On Thu, May 01, 2003 at 02:56:12PM +0100, Alan Cox wrote:
> On Iau, 2003-05-01 at 14:52, Dirk Eddelbuettel wrote:
> > Performance under 2.4.20 with ext3 + win4lin patch
> >   - "Stable" until disk-heavy operation causes freeze, typically within 
> >     two to three days
> >   - Heavy disk use (full backup writing, diff against big tarball, bonnie++) 
> >     freeze the machine hard, no ping, no sign of live
> >   + memtest86 shows no problem with the ram
> 
> If you don't have a PS/2 mouse plugged into the box add one. If that

My bad: I do have a PS/2 mouse plugged in.

> doens't help duplicate the crash without win4lin.

Good idea, will do and will try your most recent rc1-ac? as well. 

Thanks, Dirk

-- 
Don't drink and derive. Alcohol and algebra don't mix.

Re: Tyan 2466 SMP locks hard with 2.4.20 + heavy disk i/o yet runs 2.2.* without problems

[Dirk Eddelbuettel]

On Thu, May 01, 2003 at 02:56:12PM +0100, Alan Cox wrote:
> On Iau, 2003-05-01 at 14:52, Dirk Eddelbuettel wrote:
> > Performance under 2.4.20 with ext3 + win4lin patch
> >   - "Stable" until disk-heavy operation causes freeze, typically within 
> >     two to three days
> >   - Heavy disk use (full backup writing, diff against big tarball, bonnie++) 
> >     freeze the machine hard, no ping, no sign of live
> >   + memtest86 shows no problem with the ram
> 
> If you don't have a PS/2 mouse plugged into the box add one. If that
> doens't help duplicate the crash without win4lin.

Turns out that 2.4.21-rc1 is rock solid. Behaved very well without win4lin,
and is looking good with4lin (survived double bonnie++ tests). 

I also unchecked CONIG_ISA, CONFIG_PM, CONFIG_BLK_DEV_PIIX which may have
contributed to the lockups under 2.4.20.

Thanks,  Dirk

-- 
Don't drink and derive. Alcohol and algebra don't mix.

Tyan 2466 SMP lock with 2.4.21-rc1, HIGHMEM + heavy disk i/o

[Dirk Eddelbuettel]

System:
  Tyan 2466 MPX board, 2 Athlon 1800MP, bios 1.01 from March 2002
  1 gb ram, ibm + maxtor ide disks, ide cdrom + cdrw 
  ati rage 128, sb pci 128, on-board 3c905 nic, ps/2 mouse
  no other cards
  hardware configuration unaltered since June of last year

Enabling or disabling the HIGHME patch is enough to trigger/suppress
hard locking.  When HIGHMEM is enabled, locking occurs very easily when
trying to copy a (cloop-mounted) Knoppix image, or even when compiling
a kernel.

I run fairly stock kernels (from Debian kernel source packages) with the 
win4lin patch as the only external patch.  ACPI is disabled. Disk access is
not tuned in any way, DMA gets enabled by default. 

Dirk

-- 
Don't drink and derive. Alcohol and algebra don't mix.

Re: Tyan 2466 SMP lock with 2.4.21-rc1, HIGHMEM + heavy disk i/o

[Dirk Eddelbuettel]

On Sat, May 10, 2003 at 08:42:15AM -0500, Dirk Eddelbuettel wrote:
> 
> System:
>   Tyan 2466 MPX board, 2 Athlon 1800MP, bios 1.01 from March 2002
>   1 gb ram, ibm + maxtor ide disks, ide cdrom + cdrw 
>   ati rage 128, sb pci 128, on-board 3c905 nic, ps/2 mouse
>   no other cards
>   hardware configuration unaltered since June of last year
> 
> Enabling or disabling the HIGHME patch is enough to trigger/suppress
> hard locking.  When HIGHMEM is enabled, locking occurs very easily when
> trying to copy a (cloop-mounted) Knoppix image, or even when compiling
> a kernel.

As a follow-up, 2.4.21-rc2 looks fine so far. The only other change was to
enable support the proper mobo i/o chipset, which I should have done earlier.

Dirk

> I run fairly stock kernels (from Debian kernel source packages) with the 
> win4lin patch as the only external patch.  ACPI is disabled. Disk access is
> not tuned in any way, DMA gets enabled by default. 
> 
> Dirk
> 
> -- 
> Don't drink and derive. Alcohol and algebra don't mix.

-- 
Don't drink and derive. Alcohol and algebra don't mix.