Review: Servercrash with kernel SuSE 2.4.16

Review: Servercrash with kernel SuSE 2.4.16

[Oliver.Schersand]

Hi,

This is the actual result of the analys of our oracle server crashes on
linux servers with suse Enterprise Server 7 and Kernel 2.4.16

The reason for theses crashes where  the compaq remote inside and compaq
health drivers. These drivers are deliverd from compaq. On stardup of these
agents, they load binary kernel modules, which are very version sensitive.
This modules  corrupt the virtual memory management of the server on heavy
load

This shows us a main problem of Linux in datacenter environment. The
automatic guarding of the local attached storage and the hardware is very
importend in this environments. In this environment we use expensive high
performance hardware. These hardware is not  good  supported by the
standard linux kernel. The companies which sell these hardware deliver not
all features of these hardware to the community of linux.  There drivers
and guarding agents are not distributed under GPL.


Here is the statement of Compaq ( HPQ)

   The new health driver that can be run on any re-compiled kernel is due
   out in the SmartStart 5.50 timeframe (3Q02).  The reason for this is
   that it also provides support for new servers as well.
   The delivery method will be via the Compaq web site and the SmartStart
   Cd.

   The customer expectation is that it will work with any re-compiled
   kernel.  Most likely that will be true however it will be impossible for
   us to validate it with every possibility due to the open source nature
   of Linux.

   Today, customers actually can run the existing health driver on
   re-compiled kernels by evoking a "fix-up" script that will be available
   on our website.  This is not an ideal solution but it will work in a
   majority of the situations.


Thank you very much for all the help with this problem

Oliver Schersand

Re: Review: Servercrash with kernel SuSE 2.4.16

[Hans Reiser]

Oliver.Schersand@BASF-IT-Services.com wrote:

>Hi,
>
>This is the actual result of the analys of our oracle server crashes on
>linux servers with suse Enterprise Server 7 and Kernel 2.4.16
>
>The reason for theses crashes where  the compaq remote inside and compaq
>health drivers. These drivers are deliverd from compaq. On stardup of these
>agents, they load binary kernel modules, which are very version sensitive.
>This modules  corrupt the virtual memory management of the server on heavy
>load
>
>This shows us a main problem of Linux in datacenter environment. The
>automatic guarding of the local attached storage and the hardware is very
>importend in this environments. In this environment we use expensive high
>performance hardware. These hardware is not  good  supported by the
>standard linux kernel. The companies which sell these hardware deliver not
>all features of these hardware to the community of linux.  There drivers
>and guarding agents are not distributed under GPL.
>
>
>Here is the statement of Compaq ( HPQ)
>
>   The new health driver that can be run on any re-compiled kernel is due
>   out in the SmartStart 5.50 timeframe (3Q02).  The reason for this is
>   that it also provides support for new servers as well.
>   The delivery method will be via the Compaq web site and the SmartStart
>   Cd.
>
>   The customer expectation is that it will work with any re-compiled
>   kernel.  Most likely that will be true however it will be impossible for
>   us to validate it with every possibility due to the open source nature
>   of Linux.
>
Well, if they supplied the source code then customers could validate it 
and fix it.;-)

>
>   Today, customers actually can run the existing health driver on
>   re-compiled kernels by evoking a "fix-up" script that will be available
>   on our website.  This is not an ideal solution but it will work in a
>   majority of the situations.
>
>
>Thank you very much for all the help with this problem
>
>Oliver Schersand
>
>
>
>  
>

I think that any product which lacks source code is unsuited for use in 
a mission-critical Linux server application.  I think customers should 
tell companies this.  When things break, you need to be able to fix 
them, and not listen to excuses about how it is somehow the customer's 
fault that they used a version of Linux that works except for the 
vendor's code.  The paradigm has changed.....

You have my sympathies.  I will restrain myself from telling you about 
buying a Dell laptop, and having customer service tell me that they 
would not provide me any support unless I installed the original 
(windows) OS when I had a clearly hardware problem.....

I should have bought IBM and waited 4-6 weeks for their incompetent 
(incompetent only in that they take too long to ship) shipping 
department to build and send me my laptop....

Hans

Re: Review: Servercrash with kernel SuSE 2.4.16

[Alan Cox]

> This shows us a main problem of Linux in datacenter environment. The
> automatic guarding of the local attached storage and the hardware is very
> importend in this environments. In this environment we use expensive high
> performance hardware. These hardware is not  good  supported by the
> standard linux kernel. The companies which sell these hardware deliver not
> all features of these hardware to the community of linux.  There drivers
> and guarding agents are not distributed under GPL.

I would suggest you review your vendor and hardware policies. The standard
Linux i2c/smbus addons support extensive power and health monitoring for
most standards based systems.

Anyone who loads a product onto a critical datacentre system where the
vendor says "well it might work, but we don't know even with the vendor
supplied kernel" is not being terribly professional about it.

Maybe your products are in fact also supported by open source code, maybe
your choice of hardware is poor. Would you buy Win2K setups where the
vendor said "well the monitoring might work, we dont know" ?

My system temperatures, power status, disk array temperatures and disk SMART
status are all happily being logged. I have no binary modules on the server.