SUMARY of :Advfs, RAID & 2 servers

From: <arturo.espinosa_at_dot.state.fl.us>
Date: Mon, 26 Jun 2000 14:49:45 -0400

I want to thanks:

Dr.Tomas P. Blinn
Alan Pittman
alan_at_nabeth.cxo.dec.com
Chris (ruhnke_at_us.ibm.com)

It was a great help the answers given to me.

the original question was:

Hello!

I have an HSZ50 SCSI controller connected via an SCSI cable to
one DEC8200 server, and from the same SCSI card in the server
runs another extension SCSI cable to a second DEC 8200 server.
(like a daisy chain)

I'm trying without success, to see the RAID as a common volume
from both servers (I have DU 4.0E). I want to use Advfs though.

I see the RAID with the same name from both servers: rrz32 (to be
exact). And I created a file domain in both servers using the same
name and binding it to the rrz32c (I prepared the rrz32c using the
GUI program that comes with DU). Later I created a file set using
the same name in both, and binding it to the recently created file
domain.
The last step was to mount the file set as "/data" for example.

Now when I create a file in /data in server 1, and listed the content
of the same directory in the second server, and I can't see the
recently created file in the other server. I know that Both directories
in each server are talking to the RAID because I can see the LED's
activity in the same disks every time that I create or copy
something to this "/data" directories in both servers.

***************and the answers :**********************************

· First answer:

   Date: 06/26/00 08:50
    From: ruhnke_at_us.ibm.com
      To: Espinosa,Arturo
 Subject: Re: Advfs, RAID & 2 servers
------------------------------------------------------------------------------
                                                                              
                                                                              
Arturo,
                                                                              
You cannot "share" a file system in DU V4.0. You CAN share with
Tru64 V5.0 -- that's the major feature of V5 TruCluster and
improved file system access is imbedded in Tru64. You can
implement TruCluster V1.6 (a separately licensed product) in DU
V4.0, but you still do not actually "share" the file system; you
mount it via NFS and TruCluster provides a "fail over" capability that
will keep the file system available should one of the nodes fail.
             
 --CHRis
________________________________________________________
· Second answer:

   Date: 06/23/00 19:02
    From: pittman_at_oeteg.com
      To: Espinosa,Arturo
 Subject: Re: Advfs, RAID & 2 servers
------------------------------------------------------------------------------
Arturo,
  The way I understand DU and shared scsi buses such as what
you have described, you will not see the data from both system. A
volume attached to a shared scsi bus can not be mounted to two
systems at the same time. In fact, the only time I tried to do a
similar set up, I caused a ADVFS domain panic and crashed one
of my servers and corrupted the file system. I called the
Compaq support center and told me what I just told you, unless
you are running Tru-Cluster, you can not even have a entry in the
/etc/fdmns directory for the device. Tru-Cluster prevents one DU
system from trying to do a scsi reset against the device if its
mounted to another system.
                                                                               
Alan Pittman
SPX Corp
Jackson, Michigan
517-780-4108
pittman_at_oeteg.com

------------------------------------------------------------------------------------------------
· Third answer:

Date: 06/23/00 18:07
From: tcb_at_doctor.zk3.dec.com
To: Espinosa Arturo
Subject: RE: Advfs, RAID & 2 servers
--------------------------------------------------------------------------------------------------
  What you are trying to do is not supported by Tru64 UNIX the way
that you might expect it to work.
                                                        

>From the way you describe the configuration, you have an "ASE",
that is, a TruCluster Available Server Environment. In such a
setup, you can mount the file system on one server at a time (not
both) and you MUST access it from the other server through NFS.
If the primary of the servers fails (the one with the file system
mounted) there are ways in the software to cause the other server
to take over control of the file system, mount it locally, export it,
and export it to the world (including to itself, typically).
            
                                                                       
You need to find and CAREFULLY read the relevant
documentation. You will NOT succeed in getting two independent
systems to share a disk on a shared RAID controller or shared
SCSI bus; it's simply not going to ever work reliably without using
the TruCluster software.
                                                                         
Tom
                                                                         
 Dr. Thomas P. Blinn + UNIX Software Group + Compaq Computer
Corporation
  110 Spit Brook Road, MS ZKO3-2/W17 Nashua, New
Hampshire 03062-2698
   Technology Partnership Engineering Phone: (603) 884-
0646
    Internet: tpb_at_zk3.dec.com Compaq's Easynet:
alpha::tpb
     ACM Member: tpblinn_at_acm.org PC_at_Home:
tom_at_felines.mv.net

--------------------------------------------------------------------------------

· Fourth answer:

      Date: 06/23/00 17:57
    From: alan_at_nabeth.cxo.dec.com
      To: Espinosa,Arturo
 Subject: Re: Advfs, RAID & 2 servers
----------------------------------------------------------------------------
                                                                            
Correct me if I'm wrong... You're mounting the file set
from both systems for read-write access at the SAME TIME?
        
                                                                            
That's not supported until V5.0A and then only in a full
TruCluster environment (last I heard). I'd get both
file sets unmounted as quickly as possible before one
  
of the systems think the domain is inconsistent and
 decides to crash the system.
             
In this case, since there's no coordination between the
 two systems, what's been written to the log on the first
 isn't known to the other yet. Without a clue that some
blocks may have changed, the other is probably using
whatever it has cached.

END
Received on Mon Jun 26 2000 - 18:51:27 NZST

This archive was generated by hypermail 2.4.0 : Wed Nov 08 2023 - 11:53:40 NZDT