Veritas-bu

[Veritas-bu] RE: [Veritas-vx] sd.conf entries affecting what tape drives are s een (GURU's and H ACKERS please read!)

2001-10-02 17:14:28
Subject: [Veritas-bu] RE: [Veritas-vx] sd.conf entries affecting what tape drives are s een (GURU's and H ACKERS please read!)
From: mike_andres AT cnt DOT com (Mike Andres)
Date: Tue, 2 Oct 2001 16:14:28 -0500
look in /opt/JNIfcaw/technotes for a description of the various
binding/masking directives.  Use this information in conjunction with that
given in the fcaw.conf file as there is inter-play between the settings.

-mike

-----Original Message-----
From: Lumpkin, Buddy [mailto:Buddy.Lumpkin AT nordstrom DOT com]
Sent: Tuesday, October 02, 2001 3:44 PM
To: Mike Andres; 'Steve Hastings'; Lumpkin, Buddy;
veritas-bu AT mailman.eng.auburn DOT edu; 'veritas-vx AT mailman.eng.auburn DOT 
edu';
emc-l AT lists.blinky-lights DOT org
Subject: RE: [Veritas-vx] sd.conf entries affecting what tape drives are
s een (GURU's and H ACKERS please read!)


Ok, so you guys are saying that you think they sd.conf and st.conf files are
parsed by the JNI driver? That's an interesting way of looking at it. Can
you point me to a document that outlines this behavior or is this just the
way that you would guess it behaves?

--Buddy

-----Original Message-----
From: Mike Andres [mailto:mike_andres AT cnt DOT com]
Sent: Tuesday, October 02, 2001 12:56 PM
To: 'Steve Hastings'; Lumpkin, Buddy; veritas-bu AT mailman.eng.auburn DOT edu;
'veritas-vx AT mailman.eng.auburn DOT edu'; emc-l AT lists.blinky-lights DOT org
Subject: RE: [Veritas-vx] sd.conf entries affecting what tape drives are
s een (GURU's and H ACKERS please read!)


Steve, 

    If you are doing persistent binding (hopefully you are via "port = " or
"wwpn = " in your sd.conf or st.conf) how will adding a new device differ
either way?  Even if you are using wild carding in the fcaw.conf file you
will still need to add one of these directives which involves editing the sd
and st.conf's.  Also, I have never heard a JNI rep "push" one way or the
other, but I have seen storage vendors recommend not using wild carding
simply because it can have unintended side effects that will make you
scratch your head for a while.  I am a big advocate of nailing down a
configuration, especially in a SAN, where order of device discovery is not
guaranteed. Also, with the fcaw driver you can setup your system so a SAN
tape can be "hot swapped", no reboot/drvconfig required.  If you are
interested in how, let me know.

-mike

-----Original Message-----
From: Steve Hastings [mailto:stevehas AT us.ibm DOT com]
Sent: Tuesday, October 02, 2001 12:33 PM
To: Mike Andres; Lumpkin, Buddy; veritas-bu AT mailman.eng.auburn DOT edu;
'veritas-vx AT mailman.eng.auburn DOT edu'; emc-l AT lists.blinky-lights DOT org
Subject: RE: [Veritas-vx] sd.conf entries affecting what tape drives are
s een (GURU's and H ACKERS please read!)



Mike,

I am not sure you actually addressed Buddy's question, but provided a
standard process for dealing with tapes/disks.

I think the  question goes more to the relationship of st.conf and sd.conf,
and the driver.conf operation. I don't have the answer either.

Just a General statement, there is one  issue that I see with your process
below. I don't believe one size fits all, the best.
How many servers are you willing  to go and change if you have a failure.
Your method, seems to be the one that JNI is pushing, I'm not convinced its
the best from an SA's perspective
considering uptime, and availability of reboots. Consider that the
device.conf is only loaded at boot.
With wildcard binding, and controlling what is seen from the disk storage
end, allows  drvconfig, to reconfigure devices, on the fly. With the way
storage( tape& disk) increase these days, you will be rebooting evertime
you add storage.
 For tape, and backup servers I am convinced port binding is the best
answer, because of the failure of tapedrives. I don't want to have to
change every, backup server, or media server each time a drive fails, here
a reboot will be necessary, but can be scheduled, during non-emergency
times.

Certainly each sites priorities are what needs to be considered. Thats why
I don't think JNI's default solution is the best.

Steve Hastings


Mike Andres <mike_andres AT cnt DOT com>@mailman.eng.auburn.edu on 10/02/2001
06:55:41 AM

Sent by:  veritas-vx-admin AT mailman.eng.auburn DOT edu


To:   "Lumpkin, Buddy" <Buddy.Lumpkin AT nordstrom DOT com>,
      veritas-bu AT mailman.eng.auburn DOT edu,
      "'veritas-vx AT mailman.eng.auburn DOT edu'"
      <veritas-vx AT mailman.eng.auburn DOT edu>, "'sunmanager AT sunmanager 
DOT org'"
      <sunmanager AT sunmanager DOT org>, emc-l AT lists.blinky-lights DOT org
cc:
Subject:  RE: [Veritas-vx] sd.conf entries affecting what tape drives are s
      een (GURU's and H ACKERS please read!)



Buddy,

  I beleive what you are seeing is a side effect of the JNI driver.  If, in
the fcaw.conf file you have a line such as "def_hba_binding = fcaw*",
change
it.  This setting unleashes the fcaw driver to bind to any device it sees,
at any target/LUN that is outlined in any SCSI compliant driver config
file.
This inlcludes sd, st, ses, and others.  You have very little control over
how things are configured with this setting, especially in a SAN
environent.
For instance, if you wish to have a tape drive set up on a SAN at target 1,
LUN 0, on fcaw3, and you have "def_hba_binding = fcaw*" in the fcaw.conf
file, if JNI scans the ses.conf file first and a device responds at target
1, LUN 0, this controller cannot bind to the tape drive at that target/LUN
cause it is already assigned.  A better way is to set the def_hba_binding
value to "non-jni" or any string that does not match the JNI driver name.
Next, in whatever driver config file you wish to configure a device, use
the
"hba = " directive to instruct that driver instance to go and look for the
device you wish to see.

-mike

-----Original Message-----
From: Lumpkin, Buddy [mailto:Buddy.Lumpkin AT nordstrom DOT com]
Sent: Monday, October 01, 2001 1:09 PM
To: veritas-bu AT mailman.eng.auburn DOT edu;
'veritas-vx AT mailman.eng.auburn DOT edu'; 'sunmanager AT sunmanager DOT org';
emc-l AT lists.blinky-lights DOT org
Subject: [Veritas-vx] sd.conf entries affecting what tape drives are
seen (GURU's and H ACKERS please read!)


Hello All,

We have seen many times where a custom sd.conf file prevents the system
(Solaris 2.x) from seeing tape drives.

Example:

an entry commented out in sd.conf for target 3, lun 0 prevents a tape drive
from being seen at this target and LUN.

another example would be any entry that specifies a parent, hba or wwn. So
in the above instance if the line wasn't commented out but rather had a
line
that also specified something like:

name="sd" class="scsi" target=2 hba="fcaw3" lun=0;

Unless there were an entry like this for hba="fcaw0" a drive on fcaw0 at
the
same target and lun would not be seen.

It seems to me that the st driver requires the sd driver for scsi access or
at minimum relies on the sd driver to "probe" for scsi devices.

Many people have argued this point with me but I have seen this behavior
too
many times ... the sd driver definately has an affect on tape drive
visibility.

When you look at the man page for sd, it says "sd - SCSI disk and
ATAPI/SCSI
CD-ROM device driver".

Has anyone else witnessed this behaviour?

If so, does anyone have a clear understanding of how the two work together?

By what I have observed, im visualizing some kind of abstraction like this:


tape drive
--------------
Fibre(JNI)
---------------
fcaw driver
---------------
sd driver
---------------
st driver
---------------
kernel
---------------
ioctl() system calls to control tape device

Does it really work this way?

I looked around in the Solaris source code and could not find the sd or the
st driver anywhere, does anyone know where I can find them?

TIA

--Buddy


_______________________________________________
Veritas-vx maillist  -  Veritas-vx AT mailman.eng.auburn DOT edu
http://mailman.eng.auburn.edu/mailman/listinfo/veritas-vx
_______________________________________________
Veritas-vx maillist  -  Veritas-vx AT mailman.eng.auburn DOT edu
http://mailman.eng.auburn.edu/mailman/listinfo/veritas-vx