ras@rayssd.UUCP (10/07/83)
[Apologies: I realize this is rather long-winded for net.wanted, but
I would hope the discussion (if any) move to net.unix-wizards.]
We have recently been considering the possibilities of
shared (dual ported) disk drives between two UNIX VAXes for user
file systems and the possible implications of doing so, and I
wondered if anyone had considered or implemented this before.
Our environment will consist of 2 large VAX 11/780's running
4.* BSD, supporting many different projects and target machines
(based mostly on VMS VAXes and military computers). This environment
is duplicated for our Classified and Unclassified computer rooms.
What we would like to accomplish is to allow access to
either VAX by any user (via an LAN), and to allow that user to
access a common pool of disk drives by identical pathnames
regardless of which VAX is connected to his/her terminal at the
time.
By doing this we hope to avoid the condition where, if one
VAX is down, the drives would have to be reconfigured to the
other VAX's MASSBUS and the single VAX would have to mount all
the other's file systems, etc., in order to support the full user
community.
It also simplifies the administration of the two machines, since the
user community and projects are all inter-related and personnel and
management are in a constant state of flux, and assigning a set of
users to any one machine would be arbitrary, and out of date in a
month.
Some of the implications we have thought of so far are that each
VAX would need a dedicated system drive (or drives) for swap,
accounting, /etc/*, and machine-dependent programs like uucp (or
is it really only one machine?). They would also need some method for
maintaining duplicate password & group files across both
machines.
I would appreciate network discussion on the following areas
of concern, particularly if you have heard of a UNIX implementa-
tion for multi-ported drives.
1. Multi-ported drives between 2 UNIX systems:
- Is it possible with in-core filesystem info?
- Are there performance degradation issues on the drives?
- Are there other deep, dark secrets that this would
cause havoc with?
2. Star-controller and HSC-controllers from DEC for UNIX:
(These being used with DEC's CI bus architecture for
multiple VAXes using a common pool of mass-storage devices)
- Won't the same types of problems have to be resolved?
- Is DEC's new support of UNIX going to include this?
- (And then, "what if some of the machines run VMS?")
3. Distributed file systems via networks:
- Advantages and disadvantages over the other 2 choices?
(Given that redundancy and performance are most impor-
tant)
- The level of transparency to the user
- The level of transparency to the bought binary applica-
tion
I think this topic might be of general enough interest to
the net to reply by followup, but I will summarize any replies I
get via mail.
[[Please put followups in net.unix-wizards, rather that net.wanted]]
--
Ralph Shaw, {allegra, decvax!brunix, ccieng5}!rayssd!ras
Raytheon Co, Submarine Signal Div., Portsmouth, RI