[CLUE-Tech] Linux and SANs

[Jim Ockers]

gager at mho.com wrote:
> 

It's better for the list and archives if you don't top-post.  This requires
more effort on your part to go through and edit out the irrelevant parts of
the message when you reply, but it will help keep the archives of the
discussion easy to read and follow.

> So if I rebuild the system with the card installed the drives will be compiled 
> into the kernel minus the nifty failover stuff correct?

The card installed or not really has nothing to do with the nature of the
driver or whether or not it is compiled into the kernel.

In my previous message I advised you to do two things if you are using a
Red Hat system.  I'm not sure you fully read or fully understood my last 
e-mail.  I'll repeat the two things you should do here if you are going
to rebuild a kernel for a Red Hat system:

1. Use the Red Hat kernel SRPM as your build base.  (Do a 'make mrproper'
to be sure it's properly initialized.)

2. Use the Red Hat kernel .config file for your system (i686, uniprocessor
or SMP, whatever) as your config base.

If you do those two things, your rebuilt kernel will have exactly the same
attributes as the original Red Hat kernel, except perhaps a different version
name and perhaps any other changes you made.

The nifty failover stuff is not part of the QLA2200 driver in the stock or
Red Hat kernels as far as I know, and it doesn't matter if you build the 
kernel yourself or if you use Red Hat's pre-built kernels - you will not get 
the failover or multipath functionality either way.

If you feel the need to compile the driver into the kernel when you rebuild
it (I'm not sure why you would want to, since the initrd function works fine
under Red Hat), you could make that change to the .config file before you
rebuild the kernel.

Hope this helps,
Jim

> 
> RG
> 
> 
> > > 
> > > Thanks for the reply. Our software correctly identifies the SAN volumes 
> if we 
> > > are using QLogics or Emulex cards so no worries there. 
> > > I was wondering if people in production environments typically compile a 
> kernel 
> > > with the drivers for the card or if they use modprobe as we do here. My 
> hunch 
> > > is that people are compiling kernels which renders our test environment 
> invalid.
> > > Do have anyt ideas on this?
> > 
> > We use the stock Red Hat kernel drivers for our qla2200.o modules and 
> hardware.
> > We only compiled a custom kernel so we could get more than 32000 
> subdirectories
> > (the default ext[23]_link_max is 32000 in ext[23]_fs.h), since our application
> > requires that.
> > 
> > The stock kernel driver in the 2.4.9 and 2.4.18 series does not include any
> > of the nifty failover functions that you could perhaps get with a driver from
> > IBM or perhaps Qlogic.
> > 
> > If the symbols are correct you should be able to use modprobe to load either 
> > the stock kernel driver or any custom driver.  I don't understand why you have
> > a concern about not being able to use modprobe if you compile a custom kernel.
> > If you do compile a custom kernel on a Red Hat box, you would do well to start
> > with the Red Hat kernel SRPM and the .config file as provided by Red Hat for
> > your architecture and kernel version.  If you do that you should not have any
> > trouble.
> > 
> > -- 
> > Jim Ockers, P.Eng. (ockers at ockers.net)
> > Contact info: please see http://www.ockers.net/
> > _______________________________________________
> > CLUE-Tech mailing list
> > Post messages to: CLUE-Tech at clue.denver.co.us
> > Unsubscribe or manage your options: 
> http://clue.denver.co.us/mailman/listinfo/clue-tech


-- 
Jim Ockers, P.Eng. (ockers at ockers.net)
Contact info: please see http://www.ockers.net/